Improvement of Instagram Direct bot — AI logic instead of scripted bot (+ optionally auto-posting Reels)

Budget: 1105 PLN Deadline: 10 days

Hello!
We are the Business Atlas team. We have read the specifications thoroughly. The pain of merged leads is very familiar to us — when instead of a real dialogue, the bot simply guesses the step, the client sees the robot and leaves. We will fix this comprehensively.
The essence of your project hinges on three critical nodes where we have solid expertise:
First (in response to your question about architecture): we strongly recommend not building a custom backend on FastAPI from scratch just for Direct, but rather using a combination of ManyChat + Webhooks + n8n/Make as an external AI orchestrator. This will preserve the user-friendly ManyChat interface for your managers, but will completely replace rigid scripts with "brains" from the Claude API. The AI will recognize intent (price, "expensive," "I'll think about it"), will maintain exclusively the Ukrainian language, and will conduct a live dialogue, using your scripts as flexible guidelines rather than rigid frameworks.
Second — automation of auto-posting (Meta Graph API): we will link content publication (Reels, Stories, posts) with your internal tags through official media containers of Facebook/Instagram. The dynamic price will be pulled directly from Google Sheets "on the fly" without manual intervention.
Third — admin panel and roles: since you already have a bot on aiogram/FastAPI for the event project, we will integrate a multi-user panel /admin with logging and roles directly there, so as not to create unnecessary entities for your team.
We will tackle both tasks comprehensively. Our experience in this area is direct: we have already launched AI agents for e-commerce that independently guide the client from the first click to payment in chats, as well as developed large-scale content automation systems (the "Content Factory" case with parallel posting of 150+ videos on 25+ accounts across 5 social networks). Our stack: n8n, Make, Claude/OpenAI API, Meta API, Telegram Bot API.
We propose to start with the first end-to-end slice (pilot): connecting Claude to ManyChat via webhooks, uploading the logic for handling objections, and testing live dialogues using the "Price / Expensive" branch directly in Instagram Direct.
Pilot timeline (Task 1): 2–3 weeks of work. The budget for the entire project (including auto-posting and admin panel) will be fixed immediately after expert diagnostics and analysis of your current code on FastAPI.
Message me privately — send the same screenshots of conversations, and we will see where exactly the bot broke down, and agree on a time for a quick call!

Budget: 6200 PLN Deadline: 24 days

For task 1, I wouldn't rewrite the entire bot from scratch. A more sensible approach is to keep the current ManyChat, connect a webhook, and move the AI logic to an external FastAPI service. This way, we can quickly stop the loss of leads without breaking the current CRM. If we can't work properly with Instagram Direct due to Meta access or ManyChat limitations, then we will create a separate architecture through the official Meta Graph API.

We will take on both tasks. The overall estimate for Direct logic plus auto-posting is 6200 PLN and about 24 days. Task 1 separately is 7-10 days, and task 2 with Reels, stories, posts, carousels, roles, activity log, and pricing from Google Sheets is another 12-14 days. Overall, it's fine to go in stages because we first need to close the gap with leads and then calmly add publications.

Similar examples - AI and action automation https://business.ingello.com/vorfahr - close in logic to a smart handler. CRM and business processes https://business.ingello.com/forma-crm - close in roles, admin panel, and action log. Main page for Freelancehunt - https://systems-fl.ingello.com

Regarding implementation, it is important to NOT give the model the right to invent prices!! We need intent classes, a fact base, Ukrainian language by default, dialogue memory, rules for price, expensive, I will think, discounts, non-standard questions. The AI should choose the answer based on meaning, but a commercial bot without restrictions practically always starts to fantasize - beautifully, but useless for sales.

To start, we need access to ManyChat, Meta Business, Instagram, Google Sheets, and the current aiogram/FastAPI code. I will clarify two points. Are there already confirmed permissions from Meta for Instagram Graph API? And the second question - can ManyChat currently transmit the entire text of the message and the dialogue history to the webhook, or only the event of a specific step in the scenario?

Budget: 1100 PLN Deadline: 4 days

Hello! We create turnkey bots, and our profile is chatbots with real AI logic on the Claude API (we work with it every day). The pain point is clear: ManyChat guesses the step number, not the meaning, so when asked about the price, it sends a generic text and loses the lead.

Regarding your questions:

1) It's better to move the Direct logic to a separate service on aiogram/FastAPI + the official Meta Graph API, rather than fixing it on top of ManyChat through a webhook. The scenario in ManyChat breaks down; a live dialogue requires the full context of the conversation and memory, which fits perfectly into its own backend. Basic scenarios (price, discounts, "expensive," "I'll think about it") are kept as tools, but the necessary AI response is tailored to the meaning of the message. The tone is lively, the language is Ukrainian.

2) Yes, I will take on both tasks: the Direct logic and auto-posting (reels/stories/posts/carousels through the Meta API, an /admin panel with roles and logs, price from Google Sheets).

3) Regarding money and deadlines: Task 1 — 1100 PLN, 3-4 days. I will evaluate Task 2 separately after clarifying the tags and roles of the panel, aiming for 2000-3000 PLN and 5-7 days.

One question regarding the essence: should the bot ALWAYS respond in Ukrainian, or adapt to the client's language? Currently, the language "jumps" due to the randomness of templates; with AI, this issue is resolved.

I am ready to show a live demo of our AI bot so you can experience the flow before making a decision.

Petro Pankov, BotCraft Group Expand

Budget: 1100 PLN Deadline: 10 days

Diana, the problem here is not in the steps, but in the fact that the bot does not understand the intent. I would translate Direct to AI processing: determining the question about the price, "expensive," "I'll think about it," non-standard requests, and context-based responses in Ukrainian, while preserving your basic scenarios. This can be linked with ManyChat via webhook, or moved to a separate service on aiogram/FastAPI — I will suggest what is more reliable for your scheme. I work with Meta Graph API and AI integrations, I can take on both tasks and discuss the plan.

Budget: 1100 PLN Deadline: 4 days

Good day! I develop in Python, have worked on similar projects with React/Node.js, and am ready for collaboration.

Budget: 1100 PLN Deadline: 5 days

Good day! I have reviewed both tasks; the case of "leaking" leads due to strict scenarios is classic. I am ready to translate your Instagram Direct to smart AI logic and implement an auto-posting module.

Answers to your questions:

Architecture for Task 1: It is definitely better to move away from ManyChat and create a separate bot on FastAPI + Meta Graph API. ManyChat imposes strict limitations on webhook structures, charges for contacts, and often breaks context. A custom architecture on FastAPI will allow seamless integration of Instagram with your current Telegram CRM, flexible management of dialogue context in the database, and sending requests to the Claude API without workarounds.

Readiness: Yes, I will take on both tasks turnkey (both Direct and auto-posting with an admin panel).

How I will implement the project:

Task 1 (AI Direct): I will connect the Claude API (for example, Claude 3.5 Sonnet) via FastAPI. I will write a system prompt (instruction) in Ukrainian with strict rules: maintain a lively tone, handle objections ("too expensive," "I'll think about it") based on your scripts, dynamically pull the current price from your Google Sheet, and never break character. The AI will analyze the entire context of the conversation, not just the last message.

Task 2 (Auto-posting): I will implement a module for working with the Meta Graph API for publishing Reels, Stories, and posts. I will write a tag parser ([reels], [geo], etc.) and develop a multi-user /admin panel within your current Telegram bot on aiogram with role differentiation and action logging.

I have successful experience integrating LLM models into commercial chatbots and working with the capricious Meta API. Expand

Budget: 1100 PLN Deadline: 7 days

Hello. I have extensive experience in developing Telegram bots. I am ready to complete the task. Can I take a look at the Telegram bot to assess the time and cost of the task?

Budget: 1100 PLN Deadline: 7 days

Diana, hello! This is Bogdan — we are just finalizing your bot for HER ERA, so I know the task with the AI logic inside out: right now, I am removing that very "randomizer of templates," translating responses to the Claude API so that the bot understands the meaning of the question, states the price, and doesn't switch from Ukrainian to Russian. I will also take care of the autoposting through the Meta Graph API with the /admin panel and the price from the Google spreadsheet — both tasks together. I suggest keeping the Direct logic on top of the current ManyChat via webhook, without a second bot — it's faster that way. Let me know if you need Reels right away or if it can be done in the second phase? Based on that, I will calculate the exact timeline.

Budget: 1100 PLN Deadline: 5 days

Hello! I have done exactly that — bots with NLP intent classification and AI functions (not a script, but understanding the meaning). aiogram/FastAPI + Claude API is my daily stack.

Regarding your questions:
• For Task 1, I recommend using an external AI processor on top of the current ManyChat via webhook — we don't break what already works in IG Direct, but replace the "brain": incoming → meaning analysis (price / objection "expensive" / "I'll think about it" / non-standard) → response in Ukrainian based on Claude. We keep the basic scenarios as context, but the AI makes the choice. Language fix (only Ukrainian) and a correct response to a direct question about the price are paramount.
• For Task 2 (auto-posting via Meta Graph API: reels/stories/posts/carousels, tags, roles, dynamic pricing from Google Sheets), I will take it with the same block.

I am ready for both. Task 1 ≈ 4–5 days; I will clarify the total volume after access to the current flow. I will send screenshots of similar bots.
— Mykola

Budget: 1100 PLN Deadline: 7 days

Hello, Diana!

I am responding point by point:

1) Architecture: if ManyChat is already reliably handling the integration with Instagram Direct, I wouldn't rewrite it with a separate bot on aiogram/FastAPI, as that poses unnecessary risk. It's better to keep ManyChat as the transport layer and move the "brains" to an external AI processor via webhook: the message comes in, goes to its service, where the LLM (Claude/OpenAI API) actually analyzes the meaning (question about price, objection "too expensive," "I'll think about it," non-standard case) and returns a meaningful response considering the context of the dialogue, rather than a random response from templates. Plus, this allows for a confidence threshold and passing complex cases to a human - a pattern I have already used in other AI agents.

2) Task 2 (auto-posting via Meta Graph API): honestly, I don't have direct experience with Meta Graph API (Instagram) - only other people's webhook/REST integrations. I'm willing to figure it out, but if you critically need someone with ready experience specifically with this API, it's better to hire a separate specialist. I confidently take on Task 1.

3) Regarding Task 1: integrating LLM for real understanding of message meanings is what I do in my AI products (Gemini API with structured outputs and validation in promAI, RAG-bot with hybrid search in youtube_helper). Portfolio - public repos: github.com/slonce70/promAI, github.com/slonce70/youtube_helper.

Cost and timeline only for Task 1 (AI logic for Instagram Direct): ~600 PLN, 6-7 days. Expand

Budget: 1100 PLN Deadline: 14 days

Hello, Diana!

I have carefully reviewed the task. I have practical experience in developing Telegram bots, AI assistants, and management systems with artificial intelligence integration.

In particular, I created my own AI manager in Telegram that analyzes user messages, takes into account the context of the dialogue, determines the client's intent, answers questions, clarifies order details, and helps automate communication with potential clients. I also developed Telegram bots for management and notifications in my own web projects, including the BabyLand system.

Therefore, I understand your current problem well: the scripted bot selects a response based on the step number rather than the actual meaning of the message. I can modernize the existing system so that the AI:

— analyzes the content and context of the correspondence;
— recognizes questions about price, discounts, and conditions;
— correctly handles objections like "too expensive" and "I'll think about it";
— responds to non-standard questions;
— always adheres to the Ukrainian language and the specified communication style;
— does not invent prices but retrieves current data from Google Sheets;
— saves the dialogue history and forwards complex inquiries to the manager.

At the first stage, I would not rewrite the entire system from scratch. I will first audit the current ManyChat bot and check what data it sends through the webhook. If the full text of the messages is available, the optimal solution would be to connect an external AI processor on FastAPI and maintain the already functioning infrastructure. If ManyChat's limitations hinder contextual logic, I will suggest migrating to a separate service with the official Meta Graph API.

I can also take on the second part of the task:

— auto-posting Reels, Stories, posts, and carousels;
— processing control tags;
— user roles in the admin panel;
— action log;
— integration with Google Sheets;
— error control and resending publications.

I can perform both tasks comprehensively and modernize the current system without losing the already functioning functionality. Before starting, I suggest reviewing the code, the current ManyChat scheme, and examples of conversations where the bot responded incorrectly. Based on them, I will prepare test scenarios and check the new AI logic before launch.

The preliminary cost for the entire scope is 12,500 UAH.
The deadline is up to 14 days.

I am ready to start after receiving access to the current code, ManyChat, Meta Business, and examples of problematic dialogues. Expand

Budget: 1450 PLN Deadline: 14 days

Good day! Regarding your questions:

I recommend keeping ManyChat as the transport and adding a webhook → external AI processor (FastAPI + Claude API). This way, we don't break what works and can launch quickly. The bot will understand the meaning of the message (price / "expensive" / "I'll think about it" / non-standard question) and respond in live Ukrainian, pulling price figures from Google Sheets. Your basic scenarios will be saved as knowledge for the AI, not a rigid script.
I will take both tasks. Honestly, for auto-posting: Instagram (reels/stories/posts/carousels) and Facebook through the official Meta Graph API — no problem, including /admin with roles and logs. The TikTok Content Posting API requires app approval — I will do it if access is granted, which depends on TikTok, not me.
Task 1 — 8,000 UAH / 5–7 days. Task 2 — 9,000 UAH / ~7 days. Total 17,000 / 2 weeks, step by step through Safe. I will review the dialogue screenshots before starting — I will consider real customer objections.

Budget: 1100 PLN Deadline: 7 days

Hello!

I would be happy to help you replace the rigid ManyChat script with a smart AI assistant and implement auto-posting. I have extensive experience with Python (FastAPI/aiogram), Meta Graph API, and integrating LLM models (Claude, OpenAI).

Answers to your questions about architecture:

1. ManyChat vs FastAPI + Meta API:
I strongly recommend implementing this as a separate bot on FastAPI, connected directly through the official Meta Graph API (without ManyChat).
- Why? ManyChat adds unnecessary latency, severely limits working with context (dialog memory), and charges platform fees.
- On FastAPI, we can implement a full-fledged context manager (for example, saving the history of the last 10-15 messages in PostgreSQL/Redis), flexibly configure the system prompt for Claude/OpenAI so that the bot communicates in a natural language, does not invent prices, responds exclusively in Ukrainian, and clearly addresses objections like "too expensive" and "I'll think about it" based on the provided context.

2. Implementation of auto-posting (Task 2):
Yes, I can easily implement this module. We will use the Meta Graph API (endpoints /media and /media_publish) to publish Reels, Stories, and posts. I will write a simple admin panel for managing users, logging activity, and dynamically synchronizing prices from Google Sheets.

I am ready to take on both tasks together (Direct AI + Auto-posting).

The preliminary cost for the entire scope (Task 1 + 2): 1100 PLN.
Timeline: 5–7 days.

I am ready to review the current repository and discuss the details. When would it be convenient for you to chat?

Best regards,
Alexey | Python & AI Engineer Expand

Budget: 1100 PLN Deadline: 5 days

Diana, good afternoon!

I will respond to your three questions.

1. For Task 1 (AI logic for Instagram Direct), I would do it through your current ManyChat + webhook + external AI processor — this way you maintain the working infrastructure of ManyChat, and the AI (Claude API) connects as the "brain": it analyzes the meaning of the message (question about price, objection "too expensive", "I'll think about it", non-standard question) and selects the appropriate response based on the context, instead of guessing the step number. We keep the basic scenarios, but the AI selects the response, with a lively, Ukrainian tone. A separate bot on aiogram/FastAPI via Meta API is more reliable in the long run, but more expensive and takes longer; if ManyChat is otherwise satisfactory, it makes sense to start with webhook + AI.

2. Yes, I will take on both tasks. Task 2 (auto-posting reels/stories/posts/carousels via Meta Graph API, multi-user/admin panel, dynamic pricing from Google Sheets) is also within my stack.

3. Cost and timeline: Task 1 (AI logic) — approximately from 500 PLN; Task 2 (auto-posting + panel) — I will assess the volume after I look at the current ManyChat and the requirements for the panel. I will confirm the exact timeline after the screenshots of the conversations you suggested.

Experience: I work with aiogram/FastAPI, integrating Claude/OpenAI API into bots and Meta Graph API — this is my profile. I am ready to get started. Expand

Budget: 1200 PLN Deadline: 7 days

This is the same vacancy we already discussed — only now you found it on Freelancehunt (60 bids here, compared to the 1100 PLN budget originally indicated by the client). Let's format the bid according to the specific platform's format.
Comment for the bid (use our draft, slightly adapted to the "bid" format, shorter):

Hello! Task 1 is exactly what I worked on in my own project: I built an AI pipeline where instead of a rigid script, the AI analyzes the content of the message and generates a response based on the context (I used LLM for content generation, with several processing stages and quality control). The problem of "the bot guesses the step, but does not understand the meaning" is solved by replacing script logic with LLM intent classification (question about price / objection / doubt / non-standard question) + generating a response considering the context of the dialogue.
Regarding your questions:

It’s better to create a separate backend service (FastAPI) that receives webhooks from ManyChat, processes the message through the AI layer, and returns the ready response back to ManyChat — no need to rewrite the existing infrastructure.
For Task 2 (Meta Graph API) — I have experience with a similar integration with the Threads API (publishing content, working with media), but not specifically with the Instagram Graph API. I am ready to dive into it within the project or focus first on Task 1 if that is a priority for you.
Cost and timelines are below.

I can show an example of my AI pipeline architecture (repository on GitHub) — if useful, I will send the link in the correspondence. Expand

Budget: 1100 PLN Deadline: 9 days

Hello!

Your issue with the rigid script in ManyChat is completely understandable - classic linear bots often "lose" hot leads when they ask non-standard questions or want to know the price right here and now. I am ready to transition your Instagram Direct to a flexible and smart AI logic that will close this gap.
Answers to your questions about architecture:

Architecture for Task 1: It definitely makes sense to build on top of the current ManyChat via Webhook + external AI processor on FastAPI (Python). Writing a bot from scratch under the official Meta API is a time trap, as Meta will require a lengthy and strict App Review to obtain permissions for reading messages. ManyChat already has all the official accesses. We will simply connect a script on FastAPI to it, which will analyze the context through the API, handle objections ("too expensive", "I'll think about it") according to your rules, but in a lively, exclusively Ukrainian language (as specified in your terms of reference).

Choice of AI (instead of Claude): Based on personal experience in bot development, instead of the expensive Claude with its strict token limits, I suggest integrating the Gemini API (next-generation Flash and Flash-Lite models). They have a powerful Free Tier (up to 1500 free requests per day), and the response speed is significantly higher. We will send the basic text to the ultra-fast Flash-Lite, and if the client sends a photo or screenshot - the system will automatically engage the multimodal Gemini 3.5 Flash. For you, this means 0 expenses on neural networks at the launch and testing stage.

Cost and timelines: The budget you specified of 1100 PLN is quite adequate. I suggest breaking the project into two sprints through the Safe platform:

Sprint 1 (Main): Setting up the FastAPI webhook, integrating the Gemini API, writing the system prompt (risk management for objections, fixing the language, working with the price) + tests. (Timeline: 3-4 days).

Sprint 2 (Auto-posting): Setting up the Meta Graph API for Reels/Stories and pulling the current price from Google Sheets. (Timeline: 4-5 days).

I am new to this platform, so I am highly motivated to work for results and a great review. I am ready to start right now with a detailed analysis of the screenshots of conversations where the bot made mistakes, to create the perfect prompt for Gemini. Let's discuss the details in private messages! Expand

Budget: 1800 PLN Deadline: 8 days

Good day.
I am ready to take your project to work.
I can conduct an audit of your Instagram bot and fix any necessary errors.
We can also discuss auto-posting separately.
Message me privately, and we will discuss all the details and can start the implementation.

Budget: 1100 PLN Deadline: 12 days

Regarding your first question: it's better to have a separate handler on aiogram/FastAPI + the official Meta Graph API, especially since you already have a FastAPI backend; the Instagram logic will become a module alongside the Telegram CRM. ManyChat is the source of the problem: it passes the step number instead of the content of the message.

To prevent the bot from leaking leads: the price is taken only from your Google Sheet via a tool call; the model physically cannot invent a number. Basic scenarios (expensive, I’ll think about it, discounts) are embedded in the system prompt, and Claude chooses the response based on the content of the message, not the step. The language is fixed in Ukrainian at the prompt and response verification level.

A separate proposal: you mentioned screenshots of leaked conversations. Please send them; I will compile a test set from them and run the new bot on the same messages before launch, so you can see the old and new responses side by side on each leaked lead. Testing the quality of LLM responses is my main specialization; I work with the Claude API daily.

I will take on both tasks: auto-posting via the Meta Content Publishing API (reels, stories, posts, carousels with your tags), /admin with roles and action logs, dynamic pricing from the Sheet. My AI bot is in production: find auth_ai_radar_bot on Telegram; it engages in context-aware dialogue and responds based on its knowledge base.

We will discuss the details regarding Meta access and the current code in the chat.

Budget: 1100 PLN Deadline: 13 days

Hello!

I will take on both tasks together.

For task 1 — I recommend a webhook on top of the existing ManyChat: it sends messages to the FastAPI server, Claude API analyzes the content and returns a live response in Ukrainian. Basic scenarios (price, expensive, I will think) are embedded in the system prompt — the AI itself chooses the appropriate one based on the context. There is no need to rewrite ManyChat.

For task 2 — auto-posting through the official Meta Graph API, all types of content (reels, stories, post, carousel), tags, admin panel with roles, and dynamic pricing from Google Sheets.

Approximately 13-16 working days for everything together. We will discuss the exact price and details privately — I want to see screenshots of conversations where the bot leaked leads, so I can immediately incorporate the correct scenarios into the prompt.

I hope for cooperation! Expand

Budget: 1100 PLN Deadline: 7 days

Hello! I have extensive experience in developing asynchronous bots on FastAPI and integrating AI models (Claude/OpenAI API). I would be happy to take on your project.

Answering your main question regarding architecture:
It is definitely necessary to abandon ManyChat and transfer everything to your current FastAPI architecture. The workarounds with ManyChat webhooks will cause wild latency, limit control over dialogue context (Thread History), and require monthly subscriptions. Since you already have FastAPI, we will simply connect the official Meta Graph API as an additional webhook endpoint. This will be a single, free, and fully controlled infrastructure.

How I will implement the tasks:

Task 1 (AI Direct): I will integrate the Claude API (Claude 3.5 Sonnet provides the most "alive" and empathetic Ukrainian language). I will write a strict system prompt with your knowledge base (handling objections "too expensive"/"I'll think about it," tone rules). The bot will analyze the intent of the message, and the current price will be fetched on the fly via the Google Sheets API. A mix of languages is completely excluded.

Task 2 (Auto-posting): I will expand the admin panel on your FastAPI. I will implement the publication of Reels/Stories/Post through the official Meta Content Publishing API based on your tags. I will add multi-user access (roles, action logging) and integration with Google Sheets.

I will take on both tasks as a complete turnkey solution. You will receive cleanly written asynchronous code that is easy to deploy via Docker.

I am ready to discuss the details of the API contracts in the chat! Expand

Budget: 1100 PLN Deadline: 5 days

Good day!
We have experience working with Meta Graph API, ManyChat, AI assistants (Claude/OpenAI), FastAPI, and Telegram bots.
We can implement both tasks.
For Instagram Direct, we recommend not rewriting everything from scratch, but integrating the current ManyChat through Webhook + external AI processor. This will allow us to preserve existing scenarios while replacing rigid logic with context analysis: AI will determine the user's intent (price, objections, "I'll think about it," non-standard questions, etc.) and generate natural responses in Ukrainian.

We can also implement autoposting through the official Meta Graph API, with support for roles, an admin panel, logging, tags for different types of publications, and automatic fetching of the current price from Google Sheets.

After reviewing the current implementation, we will be able to propose an optimal architecture and provide an accurate estimate of costs and timelines.

Budget: 1111 PLN Deadline: 1 day

Hello! I will complete your task quickly and efficiently.

My recent works
https://indexfast.pp.ua - fast website indexing
https://mono-bank.pp.ua - everything about Monobank
https://mamamia.pp.ua - online store
https://programist.pp.ua/ua/portfolio/ - portfolio of works
https://monitortest.pp.ua - monitor testing
https://keytest.pp.ua - keyboard testing
https://pctest.pp.ua - computer testing
https://vseetut.matviy.pp.ua - layout

My portfolio: https://freelancehunt.com/ua/freelancer/romas6ka.html#portfolio
Write to me, I will start working today. I will be happy to collaborate with you! Expand

Budget: 1100 PLN Deadline: 5 days

Hello, Diana! I will move the logic of the Direct bot to a separate service on FastAPI + the official Meta Graph API instead of the fragile script in ManyChat. The bot will analyze the content of each message through the Claude API, determine the intent (price, expensive, I will think, unconventional question), and respond appropriately in live Ukrainian, maintaining the context of the dialogue - without made-up prices, only based on a database of facts. I will keep the basic scenarios with prices and objections as a foundation. I am also ready to optionally take on auto-posting Reels through the Meta API as a separate stage. How many leads is the bot currently losing per week, and are there already confirmed permissions from Meta for the Instagram Graph API?

Budget: 6000 PLN Deadline: 21 days

Good day,

I have no problem taking on both tasks. I have over 5 years of experience in developing bots on aiogram/FastAPI and integrations with AI models (OpenAI, Claude API) - exactly what is needed for Task 1.

Regarding the architecture: I recommend moving away from ManyChat and creating a separate handler on FastAPI, connected to Instagram via the official Meta Graph API. ManyChat webhook is a limited solution for AI logic, while a custom backend provides full control over the dialogue context, memory, and prompts. Claude API fits perfectly here - you can set the tone, language, basic scenarios (price, "expensive", "I'll think about it") and let the AI choose the response based on the context.

Task 2 (auto-posting via Meta Graph API - reels, stories, carousels, multi-user/admin panel, dynamic pricing from Google Sheets) will also be implemented turnkey.

The total cost for the entire scope is from $1500, with a timeline of 3-4 weeks. After launch, 2 weeks of support is free. I look forward to your message in private!

Budget: 1100 PLN Deadline: 7 days

Hello!

I did something similar for an AI consultant in Instagram Direct, where instead of rigid scripts, the bot analyzed the user's intent, qualified leads, personalized responses, and transmitted data to CRM through automation. I also built AI funnels with Chatfuel, Voiceflow, OpenAI/Claude, and Make.com, so I understand well the problem when a scripted bot loses leads due to irrelevant responses.

In your case, I wouldn't keep AI within ManyChat scripts. I see the optimal option as a webhook + external AI processor (FastAPI) that analyzes the content of the message, determines intent (price, "expensive," "I'll think about it," non-standard question), controls the single Ukrainian language, and returns a relevant response. This will preserve business logic but make the dialogue natural.

I can also implement the second part — auto-posting via Meta Graph API (Reels, Stories, posts, carousels), integration with Google Sheets for current prices, user roles, admin panel, and action log.

Approximately:

AI Direct + integration — $700–1000;
Auto-posting and admin panel — $800–1400;
Comprehensive — $1500–2200, depending on the current code and project status.

It would be helpful to look at the current implementation and examples of conversations where the bot made mistakes. After a brief audit, I can propose an architecture that minimizes lead loss and remains scalable in the future. Expand

Budget: 1090 PLN Deadline: 1 day

Hello! I have 3 years of experience in full-stack development, specializing in building ecosystems around FastAPI and integrating artificial intelligence into business processes. I have created autonomous AI agents (specifically, self-hosted solutions based on local and cloud LLMs with Docker sandboxing and n8n), so I clearly understand how to make the model not just "guess" steps but engage in deep contextual dialogue without lead leaks.

Here are specific answers to your questions and my vision for the architecture:

1. What architecture to choose for Task 1?
Definitely a separate bot on FastAPI + the official Meta Graph API.

Why not ManyChat: Pushing complex context logic, system prompts for Claude API, handling objections, and maintaining dialogue history through ManyChat webhooks creates yet another "hack" that will limit AI flexibility and cost more in maintenance.

How it will work on FastAPI: The bot fully controls the incoming webhook from Meta, passes the conversation history to Claude API (with a clear system prompt, a knowledge base about events, and scripts for handling "expensive/let me think"), analyzes the user's intent, and instantly responds in live Ukrainian.

2. Am I taking on both tasks?
Yes, I am taking on both tasks. They perfectly combine into a single infrastructure, as the same tool—Meta Graph API—is needed for both the AI assistant in Direct and for auto-posting (Reels, posts, carousels). Having an admin panel on FastAPI will allow flexible management of both manager rights and publication logs, as well as dynamic price parsing from Google Sheets (via Google Drive/Sheets API).

An important nuance regarding auto-posting: The official Meta API fully supports Reels, posts, and carousels. For automating Stories, there are strict limits from Facebook itself (native countdown or poll stickers via API are often restricted), so we will design this module considering the platform's technical capabilities to avoid account bans.

3. Preliminary cost and timelines
Since I think like an architect, I break the project into clear stages (Milestones):

Stage 1: AI logic in Direct (Task 1): Development of FastAPI backend, integration with Claude API, writing a system prompt in Ukrainian, testing context and handling objections.

Timeline: 7–10 days.

Stage 2: Auto-posting and Admin Panel (Task 2): Integration of Meta API for publications, connecting Google Sheets for dynamic pricing, creating a multi-user /admin panel with logs.

Timeline: 10–14 days.

Total timeline: ~2.5–3 weeks.

Cost estimate: I work for a fixed price for each. Expand

Budget: 1800 PLN Deadline: 5 days

Hello, Diana,

I read carefully, and the main issue is not with any broken step. The thing is that the script bot guesses the step number instead of understanding the message, so it fundamentally cannot reliably answer a direct question about the price. This is a model problem, not a configuration issue, and this is exactly what transitioning to real AI solves.

I am answering three questions:

1. Architecture. My recommendation is to start with a hybrid: keep ManyChat as the transport since it is already connected to your Instagram and approved by Meta, and add a FastAPI webhook with a Claude processor behind it. Every incoming message in Direct goes to Claude with a system prompt that knows your scenarios (price, discounts, "expensive," "I'll think about it," off-script questions), always responds in Ukrainian, and pulls the current number from the Price column in your Google Sheets, so you never have to send vague text regarding the price again. Basic scenarios are preserved, but Claude chooses the response from context, not a randomizer. This is the fastest way to stop losing leads. A completely separate bot on aiogram/FastAPI through the official Meta Instagram API is cleaner in the long run, but requires message permissions and app review in Meta, which takes real time, so I would start with a hybrid and migrate later if you want to move away from ManyChat.
2. Yes, I will take both tasks. Task 1 is a priority because it is currently costing you leads. I will do Task 2, auto-posting through the Meta Graph API Content Publishing (reels, stories, posts, carousels with your tagging system, multi-user panel /admin with roles and activity log, dynamic pricing from Sheets), after launching Task 1.
3. In phases, to quickly stop the losses: Task 1 is about 1800 PLN, approximately 5 days. Task 2 is about 2200 PLN, approximately 7 days. I will clarify both when I see the code of the current bot and the settings of your Meta account. I am ready to do Task 1 first, and regarding Task 2, you will decide after you see the result.

This is my main work: Claude and OpenAI APIs, FastAPI, API integrations. Statistics: 36 projects, all completed, rating 10/10.

One question: is your Instagram already a Business or Creator account linked to a Facebook page, and does the ManyChat connection have active message permissions? This affects how quickly the hybrid path can go live.

Revaz Expand

Budget: 1050 PLN Deadline: 7 days

Hello. Yes, I will take on both tasks together; I have experience with this.

Regarding the architecture: I recommend not leaving the logic on top of ManyChat, but rather creating a separate handler on FastAPI, connected directly to Instagram via the official Meta Graph API. ManyChat works through webhook as well, but it limits flexibility, and it is precisely due to its scenario model that problems with random templates arise. Direct access through the Meta API provides full control over the dialogue.

For task 1: the bot will analyze the content of each message through the Claude API, determine the intent (questions about price, objections about being expensive, doubts like "I'll think about it," and non-standard questions), and respond appropriately, in a natural language, consistently in Ukrainian. We will keep basic scenarios with prices and handling objections as a foundation, but the AI will choose the response based on the context.

For task 2: auto-posting Reels, stories, posts, and carousels through the Meta Graph API, an /admin panel with roles and logs, and pulling the current price from Google Sheets.

I have experience with the Meta Graph API, integrating Claude and OpenAI into chatbots, as well as with aiogram and FastAPI. I will show my portfolio with similar projects.

Budget: 1100 PLN Deadline: 7 days

Good day! I have reviewed the tasks — I am ready to help.

Task 1 — what’s better: I recommend ManyChat + webhook + external AI processor (FastAPI). ManyChat remains the channel in Instagram Direct, and messages go to your server: intent detection (price, "expensive", "I will think about it", non-standard question) → Claude or OpenAI → response essentially in Ukrainian, with a strict system prompt (prices, discounts, brand tone, no hallucinations). A separate bot only on Meta API for Direct — longer and more complicated to maintain; makes sense if ManyChat hits limits. For your case, webhook is usually faster and cheaper to implement.

Task 2: logically in the same FastAPI backend: Meta Graph API (reels, stories, posts, carousels), tags, /admin (roles, log), price from Google Sheets. With task 1 — one architecture, common Meta authorization.

Both tasks together: yes, I will take it on.

Experience: Telegram bots on aiogram, FastAPI, AI integrations, API; I do not have direct cases of ManyChat + Instagram in my portfolio — the architecture is the same as in CRM bots and AI services.

Portfolio: https://yegor10.github.io/PortFolioWeb3/ · https://github.com/yegor10 Expand

Budget: 1100 PLN Deadline: 14 days

Hello!

I have extensive experience working with the listed technologies.
You can view examples of my work in my portfolio.
I am ready to complete the assigned task.

I suggest we discuss the details in private messages.

Budget: 1100 PLN Deadline: 3 days

Hello!
I specialize in creating intelligent AI agents and automating business using ManyChat, Voiceflow, Make.com, and AI APIs (Claude/OpenAI).

Task 1
It is definitely better to work on top of the current ManyChat through Webhook + external AI processor (Make.com). Because ManyChat is the official and secure bridge with the Meta Graph API. If you write a separate bot on FastAPI for Instagram from scratch, you risk getting your account banned by Facebook's algorithms, and you will also spend a lot of time going through app moderation.
The combination of ManyChat (interface) + Make/Claude (brain) will solve your problem in a few days: the bot will communicate strictly in one language, analyze context, keep the conversation history in memory, and flexibly handle objections like "too expensive/let me think," providing exact figures.

I am ready to fully take on Task 1. I will not be able to complete Task 2. It requires pure development in Python (complex coding of the /admin panel, role settings, and direct work with the Meta Graph API), which is the domain of backend developers. However, I can set up dynamic price pulling from your Google Sheet directly into the AI bot to make your work easier.

Budget: 1500 PLN Deadline: 14 days

Good day, your problem is clear: ManyChat is currently randomizing templates and losing leads. It is necessary to transfer the dialogue brain to an external AI processor that classifies intentions, maintains context, and generates live responses in the chosen language. I can help you with both tasks.

Budget: 1100 PLN Deadline: 5 days

Hello, I have the necessary experience, I have implemented AI chats for sales and consultations, as well as AI news support and AI posters, to manage Instagram and other social media.
My Git: https://github.com/onyx144

Budget: 1100 PLN Deadline: 7 days

We have experience integrating AI logic into existing CRM systems based on FastAPI. We implement the replacement of the scripted bot in Instagram Direct with an intelligent agent that will correctly handle requests. We will also set up automation for publishing Reels via API. We are ready to start analyzing the current architecture and implementing updates.

Budget: 1100 PLN Deadline: 10 days

Good day! For over 8 years, we have specialized in developing bots with API integrations, AI models, and CRM logic, so this task is familiar to us.
Regarding the Direct bot: I would first assess the current ManyChat. If triggers and the funnel are already set up there, the fastest option would be to keep ManyChat as a shell, while moving the logic to a webhook + external AI processor on FastAPI with Claude/OpenAI. The AI will analyze the context, determine intent: price, "expensive," "I'll think about it," non-standard questions, and respond in Ukrainian in a live tone considering your scenarios and the current price from Google Sheets.
We can also handle auto-posting: Meta Graph API for posts/reels/stories/carousels, roles in /admin, activity log, tags like [reels], [stories], [geo], [countdown], pulling the price from Google Sheets.
Approximately: task 1 — 5-7 days, task 2 — 10-15 days. In total: 15-20 working days after access and clarification of Meta API limitations. I will provide the cost after reviewing the current ManyChat/CRM, as it is important to understand what has already been implemented and what access there is to Meta Business.

Budget: 1100 PLN Deadline: 4 days

Hello. I specialize in developing asynchronous services in Python (FastAPI) and integrating AI models. I propose to completely abandon builders like ManyChat and implement a reliable custom solution based on the official Meta Graph API and Claude/OpenAI API.

Answers to your questions:

1. ManyChat + Webhook or pure FastAPI + Meta API?
Definitely pure FastAPI + official Meta API.

Why this is better for you: builders (ManyChat) charge a monthly fee that increases with the number of your contacts and have limitations in transferring long dialogue history for AI. A custom bot on FastAPI will run on your server (cheap VPS), providing full control over the context and memory of the dialogue through a database. Moreover, since the Meta Graph API is still needed for Task 2 (auto-posting), we will immediately lay down the correct, unified architecture without "crutches."

Cost, deadlines, and work format:

To quickly stop the lead leakage, I suggest focusing on solving the critical problem and implementing the first task as a priority.

Stage 1: Custom AI logic for Instagram Direct

What will be done: setting up the official Webhook from Meta, deploying the FastAPI server, connecting a database to save context (so AI remembers the conversation history), integrating Claude or OpenAI API, and writing an accurate system prompt (fixing the Ukrainian language, a lively tone, clear answers to pricing questions, and flexible handling of objections like "too expensive"/"I'll think about it").

Cost: 1100 PLN.

Deadline: 4–5 working days.

Regarding Task 2 (auto-posting and admin panel) — I suggest discussing its technical requirements in detail, estimating, and implementing it as a separate stage immediately after the successful launch of the AI moderator in Direct.

My stack: Python, FastAPI, PostgreSQL/Redis (for instant context handling), Claude/OpenAI API, Docker.

I am ready to review screenshots of conversations to consider the current bot's prompt errors for AI and start working. I look forward to your feedback in private messages! Expand

Budget: 1100 PLN Deadline: 5 days

Hello!

I work with AI bots and Meta API integrations, so both tasks are within my scope. I will answer your questions right away.

Regarding the architecture of task 1: it should be done on top of the current ManyChat via webhook + external AI processor, not as a separate bot. The reason is simple — ManyChat is already legally connected to Instagram (official API), there's no need to break that and go through verification again. I will set up a FastAPI endpoint: ManyChat sends incoming messages → I pass it to the Claude API with a system prompt (basic scenarios price/"expensive"/"I'll think about it" are preserved, but the AI chooses the response based on meaning, not step number) → the response is returned to ManyChat. This addresses all three pain points: I fix the language strictly to Ukrainian in the prompt, the bot states the price in numbers (grounding — prices in the prompt/from the table), and this is a real dialogue, not a randomizer of templates.

Both tasks together — yes, I will take them on. Task 2 (auto-posting reels/stories/posts/carousels via the official Meta Graph API + tag parsers + multi-user admin panel with roles and logs + dynamic pricing from Google Sheets) — is also my area.

The total cost and timeline are below; I will confirm the final details after accessing the current bot and prompts.

We can discuss the details in private. Expand

Budget: 11000 PLN Deadline: 5 days

Good day!
I develop in Python!
I have created many bots - with AI responses, and I also have experience with Meta API.
I am ready to take on your task.

Budget: 1100 PLN Deadline: 14 days

Good day, Diana!
I have reviewed the task.

It is optimal to initially keep ManyChat as the channel but move the logic to a separate AI processor:
ManyChat → webhook → FastAPI → Claude/OpenAI → response in Direct
This way, we can quickly replace rigid steps with proper AI logic without breaking what already works.

I will take on both tasks.

To avoid delaying the launch and losing leads, I suggest splitting the work into 2 phases.
Phase 1 — AI logic for Direct: 3-5 working days, budget to be estimated during our conversation
Phase 2 — auto-posting + roles + admin panel + activity log: after checking Meta API permissions, 5-7 working days, budget to be estimated during our conversation

So, for a complete project assessment, it is enough to see the current ManyChat flow, 2-3 examples of problematic conversations, and the structure of the Google Sheet with prices.

I am ready to start with an audit and immediately show the scheme of how AI will determine intents and formulate responses.

Official partner of well-known chatbot platforms

I await your message for discussing the terms and budget of the project.

Portfolio
Freelancehunt Expand

Budget: 1100 PLN Deadline: 7 days

Hello, I have already worked with both the official and unofficial Instagram API, and I also have experience connecting AI to various bots. If you are interested in quality task execution, please message me privately to discuss the details and we will do everything in the best way.

Budget: 4900 PLN Deadline: 19 days

Hello. Our team specializes in integrating AI assistants into sales departments and working with the Meta Graph API. You are losing leads because your current bot has been functioning as an auto-responder since 2018, rather than as a salesperson. We will handle both of your tasks turnkey.

Answers to your questions and our Architecture:
1. Direct (Regarding ManyChat vs Custom Bot): The best solution for business is hybrid. We keep ManyChat as a convenient gateway (so your managers can comfortably intercept dialogues through Live Chat). But the "brain" becomes our external processor on FastAPI. ManyChat sends messages to the server via Webhook -> AI (Claude 3.5 / GPT-4o) analyzes the context, retrieves the relevant price from your Google Sheet, and returns the perfect response, addressing the objection "too expensive."
2. Auto-posting (Task 2): We take this 100%. On the same FastAPI server, we will deploy a module to work with the Meta Graph API. We will implement a multi-user admin panel directly in your Telegram bot for convenient posting of Reels/Stories/Carousels with tags.

Regarding cases: the internal AI algorithms and sales funnels of our clients are strictly protected by NDA (commercial secret), so I do not share others' data. However, I propose a more transparent approach:
Write to me, and I will specifically create a basic test processor for your services. You will be able to check live how the AI responds to your objection "too expensive" and leads the client to dialogue, even before we start discussing financing details.

Budget: 1100 PLN Deadline: 4 days

Hello!

I have reviewed your order for the improvement of the bot for Instagram Direct and auto-posting. I am ready to take on the implementation.

How is it best to approach Task 1 (AI logic)? It is best to create a separate bot using aiogram/FastAPI through the official Meta Graph API. The ManyChat + webhooks architecture has strict limits, cuts off context, and complicates the transfer of dialogue history to LLM (Claude/OpenAI). Clean code in Python will provide complete freedom in managing context, system prompts, and will reduce latency in AI responses.

Budget: 1100 PLN Deadline: 2 days

Hello. I am ready to take on your order. I have extensive experience in implementing AI, in creating AI chatbots. I can also implement automatic uploading of reels using an agent system. I can start today. I will finish quickly.

Budget: 1100 PLN Deadline: 4 days

Hello! The topic is close — I am developing AI logic for chatbots using Claude/OpenAI API, so right away regarding your questions:

1. Architecture. The problem with the current bot is that ManyChat is script-based by nature, and a "real" AI dialogue always runs into its limitations. The optimal solution: keep ManyChat as the reception channel, but each message goes through a webhook → external service (FastAPI + Claude API), which analyzes the meaning (question about price / "expensive" / "I'll think about it" / off-topic) and returns a response. We keep basic scenarios as "frameworks," and the AI formulates responses based on the context. If you want full control and to move away from ManyChat's limits — I can transfer it to a separate bot using aiogram/FastAPI on the official Meta Graph API. I will recommend the final solution after reviewing the current setup.

2. Both tasks — yes, I will take on the entire scope (Direct logic + auto-posting). It is more convenient for me to handle this with one stack.

3. What I will essentially cover:
• The bot understands the meaning of the message, rather than guessing — it provides a number in response to a direct question about the price (including dynamic data from Google Sheets)
• I fix the "jumping language" — I strictly set it to Ukrainian
• Dialogue context (remembers what the conversation is about), a lively tone, without sounding "robotic"
• Auto-posting through Meta Graph API: reels/stories/posts/carousels, parsing tags [reels]/[stories]/[post]/[carousel]/[poll]/[countdown]
• Multi-user / admin panel: adding/removing people, roles (posting / viewing activity)

Experience: integration of Claude/OpenAI API into bots, working with webhooks and Meta Graph API.

I am ready to have a 10-minute call to review the current bot and screenshots of conversations — after that, I will provide an exact final deadline and cost for each task. Preliminary estimate — in the bid. Expand

Budget: 1500 PLN Deadline: 5 days

Hello.

Yes, the task suits us. For the Direct bot, I would initially do not a full transfer, but an improvement of the current ManyChat through webhook + external AI processor. This is faster, cheaper, and will allow you to keep your current scenarios while adding proper understanding of the meaning of messages.

The AI will be able to determine: questions about price, "expensive," "I'll think about it," discounts, non-standard questions — and respond contextually in Ukrainian, without random templates.

We can also discuss auto-posting through the Meta Graph API as a second stage: reels, stories, posts, carousels, roles in the admin panel, logs, and pulling prices from Google Sheets.

I suggest starting with an audit of the current ManyChat bot and conversations, after which we can provide an accurate estimate of the timeline and cost.

Budget: 1100 PLN Deadline: 7 days

Hello, Diana!

I completely understand your problem: rigid ManyChat scenarios with random text selection often break the logic of the dialogue, annoy customers, and consequently lead to the loss of hot leads. I am ready to help you elevate communication to a qualitatively new level through the integration of the Claude API.

I am responding to your questions regarding the architecture:
1. The optimal solution for Task No. 1 is to keep ManyChat as a stable message delivery interface but connect an external AI processor via Webhook based on FastAPI. On the server, messages will be analyzed by the Claude API with a clear system prompt (where all triggers will be specified: price, "expensive," "I'll think about it," language rules, and tone of voice), while the dialogue history (context) will be stored in the database. This is significantly faster to implement and will not break what is already working. If we hit limitations, then we will move Direct to a separate service via Meta API.
2. Yes, I am ready to take on both tasks comprehensively (smart Direct + auto-posting Reels/Stories with an admin panel in Telegram).
3. I am ready to agree on the exact cost and timelines after reviewing examples of problematic conversations and the current structure of ManyChat. Approximately for the entire scope — according to your budget, in terms of timelines — up to 7-10 days (work in iterations: first, we save leads in Direct, then we set up auto-posting).

In freelancing, I specialize in Python (FastAPI, aiogram 3.x), and I deploy all my projects on autonomous hosting Railway (operating 24/7). You can view examples of my working bots with databases and integrated payments (LiqPay) in my portfolio.

I would be happy to collaborate. Please let me know if the screenshots of the problematic dialogues are available so that I can analyze the logic in more detail before starting? Expand

Budget: 1050 PLN Deadline: 7 days

Good morning,

I can do this. It's best to start with task 1: keep the current ManyChat and add a webhook + external AI-handler in FastAPI, so the bot analyzes the meaning of the messages instead of following rigid steps.

I will improve intent recognition: price, "expensive," "I'll think about it," custom questions, Ukrainian language, fallback, and handover of the conversation to a human.

I can also do autoposting via the Meta Graph API, but as a second stage after checking access and API limitations.

For stage 1, I propose 950 PLN / 7 days.
I can estimate the total with autoposting after a brief discussion of the details.

Budget: 1100 PLN Deadline: 4 days

Good day! I have reviewed the task — I will take on both tasks.
For task 1 — I recommend using a webhook over ManyChat with an external AI processor on FastAPI + Claude API. This is faster than rewriting everything from scratch, and ManyChat remains as the interface. The bot will truly understand the content of the message — questions about price, objections like "too expensive," "I'll think about it" — and respond appropriately in Ukrainian without language jumps.
For task 2 — auto-posting via the Meta Graph API is feasible, all types of content are supported. A multi-user panel with roles and activity logs, dynamic pricing from Google Sheets — everything is included.
Please send the screenshots of the problematic conversations you mentioned — it will be immediately clear where the logic breaks and what exactly needs to be fixed. I am ready to discuss the details.

Budget: 1100 PLN Deadline: 6 days

Hello!

I am ready to implement both tasks turnkey.

Regarding the smart bot in Instagram Direct (Task 1):
I propose to create a system from scratch without using ManyChat to achieve maximum reliability. The bot will be directly connected to Instagram, will analyze the content of messages, and will respond in live Ukrainian. For text processing, I suggest using DeepSeek — it demonstrates excellent context understanding and is significantly more cost-effective than other models. The bot will automatically pull the current prices from your spreadsheet.

Regarding autoposting and Telegram admin panel (Task 2):
I will create a convenient management panel in Telegram. You will be able to manage the team, and the bot will automatically publish posts, reels, and stories, taking fresh information from Google Sheets.

I am always in touch, maintain clear communication, and show initiative to ensure everything works perfectly. I am currently actively building my rating on this platform, so I offer the most favorable conditions.

Cost for the entire project: 1100 PLN
Timeline: 5-6 days.

I am ready to discuss the details and start working. Message me! Expand

Budget: 1100 PLN Deadline: 7 days

Hello! I understood the task — to bring the Instagram Direct bot from "guess the step" to a normal live dialogue on AI, plus to implement auto-posting of Reels/Stories through the Meta Graph API.

For task #1 (Direct logic): ManyChat with its randomizer of templates really loses leads because it doesn't understand the context. I would go for a separate bot on aiogram/FastAPI, connected to Instagram through the official Meta API — the webhook receives the message, Claude API analyzes the meaning (question about price / "expensive" / "I'll think about it" / non-standard), and the bot responds essentially in Ukrainian, maintaining the specified tone. Rigid basic scenarios will remain as insurance — but AI will choose the appropriate response based on the context, rather than firing random text.

For task #2 (auto-posting): through the Meta Graph API, I will implement the publication of reels/stories/posts/carousels with tags [reels], [stories], [post], [carousel] + geo/polls/countdown options. An admin panel /admin with roles (posting/viewing/all) and an activity log, dynamic pricing from Google Sheets (column "Price") will be pulled into the bot's messages. I will take on both tasks together — doing it comprehensively.

⚙️ Stack: Python + FastAPI for the webhook handler, Claude API/OpenAI for dialogue semantics, Meta Graph API for Instagram Direct and posting, Google Sheets API for current prices. I already have experience integrating AI models into chatbots and working with the Instagram Graph API.

Examples of similar work:
AI bot for CRM event project on aiogram/FastAPI — intent analysis of leads through OpenAI API, dynamic responses instead of rigid scenarios. Conversion to qualified leads increased from ~12% to ~31%.
Auto-posting bot for a clothing brand on Instagram (Meta Graph API) — publications of reels/stories/carousels on schedule, multi-user admin panel with roles and action log.

I will take on both tasks — starting today. Please reply with a few words about your preferences regarding the Claude/OpenAI stack — and I will draft a plan for the first stage. Expand

Budget: 1100 PLN Deadline: 7 days

Hello, Diana! 👋
I have reviewed the specification — the task is clear and I can take on both points.
Task 1 — a classic problem: a rigid ManyChat scenario is not suitable for live dialogue. The best solution is to leave ManyChat as a messaging delivery channel but connect an external AI handler via webhook: ManyChat receives a message → sends it to the FastAPI server → there Claude API analyzes the context and returns a response back to Direct. This is quicker to implement, does not require rewriting the infrastructure, and ManyChat remains a convenient UI for management. An alternative — a full-fledged bot through Meta Webhooks + aiogram/FastAPI, but this takes longer and is more expensive for the same effects.
Task 2 — I have implemented autoposting through the Meta Graph API, so carousels, reels, stories, geotags — no problem. Retrieving the current price from Google Sheets via gspread is also a familiar issue to me.
The stack I use: FastAPI, Claude API / OpenAI API, Meta Graph API, Google Sheets (gspread), Docker.
Details regarding costs and deadlines — in a private message.

Budget: 14500 PLN Deadline: 24 days

We already have a practically ready similar solution for an AI bot for dialogues and lead processing, which can be quickly adapted for Instagram Direct and your CRM (:
I'm available, we can discuss the details here on the marketplace.

Regarding the architecture, I would start with ManyChat webhook plus an external AI processor on FastAPI - this is faster for task 1 and less disruptive to the current system.
If we run into limitations with ManyChat or Meta rights, then we can move Direct to a separate service through the official Meta API.
It's important not just to connect Claude API or OpenAI API, but to build a layer of rules, memory, and response verification so that the AI doesn't invent prices and doesn't confuse languages.

We will take on both tasks.
Task 1 - AI logic for Direct, Ukrainian language, intents for price, expensive, I will think, non-standard questions, dialogue context.
Task 2 - auto-posting reels, stories, posts, carousels, tags, roles in /admin, activity log, pulling prices from Google Sheets.
Overall estimate for the entire volume - 14,500 PLN, timeline - about 24 working days.
The initial phase for Direct can be highlighted separately - 5,500 PLN and 7-10 working days.

Based on experience, we are close to your task:
- https://business.ingello.com/vorfahr - AI automation and agent logic for business processes
- https://business.ingello.com/fractal - complex automation with agents and workflows
- https://business.ingello.com/forma-crm - CRM and operating system with roles, logic, and admin panel

Main page for projects from the marketplace - https://systems-fl.ingello.com

We will need access to ManyChat, Meta Business, Instagram professional account, current Telegram bot or repository, Google Sheets, and examples of conversations where the bot responded off-topic.

I would like to clarify 2 points.
- Is Instagram already connected as a professional account to Meta Business, and are there confirmed rights for messaging via API?
- Is auto-posting needed only for your accounts, or will there be different clients and different Instagram accounts in the panel?

Overall, it feels better to approach the task in stages - first save Direct and leads, then carefully add posting.
Not all that glitters is gold, and not every bot with a model becomes a sales tool.)) Expand

Budget: 1100 PLN Deadline: 10 days

ManyChat currently does not understand the meaning — it just guesses the step, so the responses are off-topic. I will do the following:

1. Webhook handler on FastAPI: ManyChat sends a message, the server forwards it to the Claude API with a system prompt (tone, language — Ukrainian, basic scenarios: price, "expensive", "I'll think about it", non-standard). Claude returns a live response, which ManyChat sends. This way, ManyChat remains the front end, but the logic is AI.

2. For auto-posting — Meta Graph API: Reels, Stories, posts, carousels via /media + /media_publish. Tags are parsed from the text in the bot, and the price is pulled from the Google Sheets API. Admin panel in Telegram: roles, action log.

3. I store the dialogue context in Redis (the last N messages) — without this, AI does not see what has already been said and responds as if it were the first time.

Please send screenshots of merged leads and the current ManyChat flow — I want to see exactly where the bot breaks, so I can accurately describe the system prompt for your scenarios.

Budget: 1100 PLN Deadline: 7 days

Hello, I see that your Instagram Direct bot is currently functioning as a template randomizer rather than a real conversational partner. The main issue is that it does not understand the essence of the clients' messages, which leads to "leaking" live leads.

Instead of rigid scripts, I will integrate an AI model (for example, Claude API) that will analyze inquiries and understand when a client asks about the price, has doubts, or expresses objections. The bot will respond in fluent Ukrainian, providing relevant information and maintaining the context of the dialogue. The result will be not just a bot, but an effective assistant that truly communicates and converts interested leads.

To start the work, I would like to clarify: is there already a ready prompt or examples of dialogues that can be used to train the AI?