Analysis of dialogues and creation of a prompt for the AI sales manager

Budget: 15000 UAH Deadline: 7 days

I can take your order. Please clarify:
1. What is the approximate average size of one dialogue and the total volume of data in JSON?
2. Do you need to generate only the final prompt_id or also separately structure the knowledge base, scenarios, and escalation rules?
3. Which OpenAI models are you currently using in the SaaS platform?
4. Should the result be generated once when a new client connects or automatically updated based on new dialogues?
5. Are there labeled examples of successful/unsuccessful dialogues, or do they also need to be determined automatically?

Budget: 27000 UAH Deadline: 7 days

The budget of 1000 UAH for this task feels unrealistic. For 30,000 UAH, we can complete the first technical stage in 7 days - architecture, prototype for analyzing JSON dialogues, generating a knowledge structure, and a draft developer message for one business =)

If we go for industrial implementation, I would build not one large request to the model, but a pipeline - cleaning dialogues, clustering intents, extracting facts, scenarios, stop situations for a live manager, examples of successful and unsuccessful dialogues, then assembling the prompt and testing on control dialogues. For this, we can use the OpenAI Batch API or Responses API, and for large datasets, add an intermediate storage and quality assessment of the results.

There is a nuance - without a quality check stage, the system can beautifully assemble a prompt but miss important sales rules. Therefore, we need not just a text generator, but a mechanism for extracting knowledge with evidence from dialogues and tests on typical situations.

> Clarification 1 - is there already a markup in the JSON indicating who wrote - client, manager, bot, or should this be determined separately?
> Clarification 2 - should the final result be only the developer message, or should there also be a separate knowledge base, escalation rules, and examples for testing?

Relevant examples from Ingello
> https://business.ingello.com/vorfahr - AI and process automation in product logic
> https://business.ingello.com/fractal - agent approaches and complex automations
> https://business.ingello.com/platforma - platform development with business logic

Main profile for Freelancehunt - https://systems-fl.ingello.com/ua

We can keep the start simple - first, we take a part of real dialogues, create a prototype for knowledge extraction, and compare the result with what a human currently creates. Then we can see where further development is needed and what the budget for full integration will be. Expand

Budget: 10000 UAH Deadline: 20 days

Good day, Oleksandr!

I specialize in the development and design of AI solutions and SaaS platforms, focusing on the automation of business processes that are currently done manually. I work with LLMs (OpenAI / Google / Microsoft) and architectures where artificial intelligence independently analyzes data and forms workflows for the product.

In order to formulate an objective proposal for implementation, I would like to clarify a few points:

- Is the structure of what "successful" prompts look like in production already defined?
- Do you have any markup or at least partial classification of dialogues (successful / unsuccessful / sales / support)?
- What infrastructure is currently used for data processing (backend, queues, storage)?
- Is this module planned as a separate service within the SaaS or as part of the existing backend?

After your responses, I will be able to accurately assess the complexity of the implementation, propose a solution architecture, and provide estimated timelines/costs.

I would suggest implementing this as a separate AI pipeline within your SaaS platform, where dialogues undergo stages of processing, structuring, and prompt generation without human involvement. Essentially, this will be a system that transforms raw customer chats into ready-made "behavior logic" for the AI manager, which can be immediately used in production.

Oleksandr, I propose discussing the details in private messages so that I can offer a more precise architecture for your current platform and show how this can be implemented without manual work from the team and constant prompt adjustments. Expand

Budget: 1000 UAH Deadline: 30 days

I am implementing an AI pipeline for analyzing JSON arrays through the Batch API from OpenAI or Google Vertex AI for distilling dialogues, automatically highlighting edge cases for transferring chat to a human, and generating a system prompt.

Are you planning to use a vector database (RAG) for dynamically connecting the found knowledge base, or should all collected information be directly baked into the final prompt_id, limiting the contextual window?

We can discuss the budget and timelines in private correspondence.

Budget: 1000 UAH Deadline: 7 days

Hello, I worked on a sales dialogue analysis system for an e-commerce platform with over 15,000 messages, where I automated the creation of AI prompts and increased conversion by 23%.

I'm curious how you plan to handle the context of long dialogues and whether you need to consider the emotional tone of customers when forming prompts?

I suggest we get in touch; I will provide you with free technical consultation and we can create a development plan together + I will tell you about my team!

Budget: 27000 UAH Deadline: 7 days

We already have a practically ready solution for such a task - it can be quickly adapted to your SaaS and launch the first working version ))

Regarding the budget of 1000 UAH, well, there’s a nuance - this will rather be enough for a consultation or a brief analysis of the approach, rather than for the implementation of the mechanism for analyzing 3000 dialogues and generating a prompt.

For the first stage, I would suggest creating a prototype in 7 days - analyzing JSON, extracting the knowledge base, scenarios, reasons for transferring to a live manager, examples of successful and unsuccessful dialogues, as well as assembling a ready developer message for OpenAI.

I see the implementation as a pipeline of several steps - cleaning dialogues, clustering topics, extracting business rules, separate analysis of sales patterns, checking for contradictions, then generating the prompt and testing on a control sample.

It’s important not just to ask the model to read all dialogues, because it will generate beautiful text and go have tea, but to build a repeatable process with quality assessment.

We have worked on similar tasks related to AI, automation, and SaaS.
- https://business.ingello.com/vorfahr - similar in logic to AI automation and business scenarios
- https://business.ingello.com/fractal - about agent processes and complex automated chains
- https://business.ingello.com/tts - an example of working with AI communication and voice scenarios

The main page for our approach for the marketplace - https://systems-fl.ingello.com/ua

What we will need from you - an example JSON, preferably 50-100 dialogues for testing, the current prompt_id or its text, a description of the final prompt format, and the criteria by which you consider the AI manager's response successful.

I would like to clarify two points.
- Do you need to generate only the prompt, or also automatically test it on historical dialogues?
- Are your dialogues already labeled as successful or unsuccessful, or does this also need to be determined automatically?

We can discuss the details here and quickly move to the prototype - I’m available (:. Expand

Budget: 11111 UAH Deadline: 1 day

Good day, Oleksandr.
Strong task formulation!

I have worked with OpenAI API, building AI assistants and chat analysis systems.
For implementation, I would look towards:
➡️ OpenAI API
➡️ RAG for knowledge extraction
➡️ Gemini 2.5 / GPT-4.1 for semantic analyses

I can assist with both architecture and MVP implementation.
Contact me, and I will provide a project estimate.

Thank you, have a nice day. Expand

Budget: 1200 UAH Deadline: 3 days

Very interesting task, especially the idea of transforming "live" Instagram/Telegram dialogues into a ready prompt for an AI sales manager without the involvement of the business owner. The key here is not just to generate text, but to correctly extract patterns: knowledge base, sales scenarios, triggers for escalation to a human, and unsuccessful cases.

I have worked with AI consultants for Instagram Direct, automating communication and prompt engineering under sales/funnel logic. For such a system, I would look towards OpenAI + structured extraction pipeline: first, classification of dialogues, then extraction of intent/scenario patterns, and after that — generation of the final developer prompt through a predefined framework.

It is also important to separately process "bad dialogs," as they often show where AI should not improvise.

I can help design the architecture of this mechanism, choose AI tools, and implement the extraction/generation pipeline for your SaaS platform.

Budget: 22000 UAH Deadline: 12 days

Good day!

Pipeline in Python: (1) clustering 3000 dialogues by topic/intent using text-embedding-3-large + UMAP/HDBSCAN, (2) knowledge extraction via Claude Opus 4.7 with structured output across 4 categories (knowledge base / scenarios / escalation situations / good vs bad examples), (3) synthesis of a system prompt through meta-prompting on GPT-5.5 with grounding on extracted patterns, (4) validation on a holdout of 10% of dialogues - checking if the new prompt reproduces real responses from managers.

Ready-to-use tools: OpenAI Evals (native prompt validation) + Microsoft PromptWizard (auto-optimization). There is no ready-made SaaS that does exactly this with 3000 IG dialogues on the market - a custom solution is needed, but the architecture is clearly defined.

A week ago, I took 3rd place solo at the AI Agent Olympics Hackathon Milan AI Week 2026 (the largest AI event in Europe, 731 teams, 2382 participants). Full-time AI engineer for over 1 year. MSc in Strategic PM, PRINCE2 - structure and documentation in every project.

Price: 22,000-30,000 UAH depending on the number of product verticals, duration 12-16 days.

Budget: 2000 UAH Deadline: 1 day

Hello! We are a team of developers and social media communication specialists with 4 years of practical experience. Creating an effective prompt based on real dialogues requires not just technical writing of instructions, but a deep understanding of client psychology and SMM practices. We will conduct a full audit of the provided chats, structure typical user requests, and develop a flexible prompt that will allow the AI agent to close clients on the target action as effectively as a top manager does. We will pay special attention to data security and exclude scenarios where the client may confuse the AI. Let's discuss the scope of dialogues for analysis in private messages!

Budget: 2000 UAH Deadline: 1 day

Hello.

I can analyze dialogues and create prompts. Write to me, and we will discuss.

Budget: 22000 UAH Deadline: 7 days

Hello! I understand the task - we need to build a pipeline that automatically generates a structured prompt for the AI manager from raw dialogues.

My approach:
- Chunking and vectorization of dialogues (OpenAI Embeddings + clustering) to identify patterns: FAQs, scenarios, stop situations
- GPT-4o with structured outputs for extracting knowledge bases, scripts, and examples in JSON
- Automatic formation of the final prompt_id through OpenAI Prompt Management or saving as a developer message

I have implemented similar things: RAG systems, chat analysis, prompt automation for business. I can show an example of a pipeline using your test data.

I am ready to discuss the details - please let me know how many dialogues on average and what the current JSON output format is.

Budget: 15000 UAH Deadline: 7 days

Good day, Oleksandr!

Your task is the pipeline "raw dialogue corpus → structured insights → ready prompt": to automatically extract knowledge from 3000 dialogues instead of "reading them with your eyes" and compile a developer message for a specific business.

Here’s how I see it (using ready-made tools, as you want):

- preprocess JSON dialogues + AI markup for each: product type, outcome (success/failure), whether there was a transition to a live manager;
- map-reduce extraction (3000 dialogues won’t fit into one context): we pull business facts in batches → dedup → consolidated knowledge base (delivery, payment, details, schedule);
- clustering by product types and situations (embeddings) → typical conversation scenario for each cluster;
- detection of patterns where a live manager is needed (complaints, complex and legal issues); selection of benchmark successful dialogues and characteristic unsuccessful ones as anti-examples;
- compiling everything into a structured prompt according to the template, ready to fit into your prompt_id.

The reliability of extraction is ensured by structured outputs (responses strictly according to the JSON schema), and the Batch API reduces the cost of processing 3000 dialogues by half.

This is exactly the architecture I constantly build: my multi-agent pipelines digest large "dirty" inputs (for example, an entire codebase), extract structured findings, and compile a final structured document. Here, the input is dialogues, and the output is a prompt, the mechanics are the same.

I suggest starting with a paid pilot: I will run a sample (~200–300 dialogues from one business) and provide a ready draft prompt + knowledge base for you to assess the quality on a small scale, then we can scale up to the full volume and other stores. The estimate: 15,000 UAH / 7 days for the pilot, full volume after that.

P.S. By the way — generating the prompt and testing it is essentially one cycle. If needed, I can also handle the automatic testing of dialogues, so everything is managed by one person. Expand

Budget: 1000 UAH Deadline: 3 days

Hello! I have carefully reviewed your project and am ready to start working. I guarantee quality and timely execution.

Budget: 1000 UAH Deadline: 1 day

Of course, I have experience in developing such solutions. I use GPT-4/OpenAI API, LangChain/LlamaIndex to process 3000 dialogues: extracting a knowledge base, scenarios, criteria for transitioning to a live manager, analyzing successful and unsuccessful chats. After preparing the vector storage, I generate a prompt that includes all key scenarios and instructions. The result is a ready prompt_id that can be imported into your SaaS platform without involving the owner.