Consultation on creating an AI agent / chatbot in Telegram

Budget: 700 PLN Deadline: 2 days

Hello! I have been building RAG bots in Telegram that respond strictly from the knowledge base and pass on to the curator when there is a gap. I will take a look at your settings in suvvy.ai and let you know if the cost of requests is reasonable and where it can actually be reduced. One immediate question: what model is the bot currently using and how large is the knowledge base? This will determine where to cut costs. I will conduct an audit with recommendations in a couple of days.

Budget: 220 PLN Deadline: 2 days

The price for such an audit is 220 USD, the timeframe is 1-2 days. The format will include an analysis of the current settings of Suvvy.ai, the structure of the knowledge base, the cost of responses, the scenario for transferring the model to the curator, and I will provide a short written conclusion with practical corrections.

We can keep it simple - the main risk here is not in the Telegram bot itself, but in how the knowledge base is structured, how the bot searches for answers, and where it starts to make things up instead of transferring to the curator. I will also check if it makes sense to stay on Suvvy.ai or if it would be cheaper to build a more manageable scheme on another platform.

I would like to clarify two points:
> Do you have access to expense statistics, dialogue logs, and model settings in Suvvy.ai?
> Is the knowledge base currently more in the form of documents or in the form of question-answer pairs?

Similar types of case studies:
> https://business.ingello.com/vorfahr - AI and SaaS, where the cost of operation and manageability of the result are important
> https://business.ingello.com/fractal - automation with AI agents and action control
> https://systems-fl.ingello.com - main profile of Ingello Systems for FLH

Note that after the audit, there will be not just a general opinion, but a list of specific solutions regarding the knowledge base, settings, and optimization of response costs. Overall, it's fine to start with such a short phase, and then we can decide whether a transfer or further development is needed =) Expand

Budget: 150 PLN Deadline: 1 day

Hello!

I would be happy to consult you on optimizing the bot and help set up a knowledge base for maximum efficiency. To give you an immediate understanding of my approach, I will answer your main questions right here:

1. Cost Optimization:
Constructors like Suvvy.ai take a large markup on top of the cost of the OpenAI API. If you transfer the logic to your own lightweight script (for example, in Python / FastAPI / aiogram), you will only pay for the "raw" tokens directly to OpenAI. Using a fast and cheap model like gpt-4o-mini, the cost of one response will literally be a fraction of a cent. Semantic caching can also be implemented — if a client asks something that has been asked before, the bot will respond instantly from the cache with zero API costs.

2. Knowledge Base Formatting:
Uploading entire text documents (PDF/Word) is inefficient and expensive. With RAG search (when AI searches for information in documents), a lot of unnecessary junk gets into the model's context, inflating the price for each token and provoking AI "hallucinations." The "Question-Answer" (Q&A) format is the gold standard for service bots. A vector database finds semantic similarity between the client's question and your prepared Q&A block much more accurately. It is best to structure the knowledge base in JSON/CSV format with a clear separation into question-answer pairs.

I can conduct an audit of your current settings, analyze the architecture in detail, advise on limits, and help transition the bot to a custom solution if you decide to move away from the constructor for cost savings.

I suggest we have a call or communicate in a format that is convenient for you. What consultation fee is comfortable for you?

Best regards,
Alexey | AI Integration Specialist Expand

Budget: 250 PLN Deadline: 2 days

I understood the task: you already have a bot working on Suvvy.ai, and you need an audit of the settings — whether the cost of responses is optimal and how to properly structure the knowledge base so that the bot responds more accurately and cheaply.

I have worked with RAG search on knowledge bases and prompt engineering on a voice AI assistant (retrieval through a vector database, fine-tuning prompts for specific dialogue scenarios) — the principles are the same as with Suvvy: the cleaner the chunks of the knowledge base are cut and the more precisely the prompt restricts the model to respond only based on the found context, the fewer tokens are used for each response and the less often the bot invents unnecessary information.

I can review the current settings of Suvvy (prompt, knowledge base format, response length) and provide specific recommendations: how to reformat documents for question-answer instead of whole files, where to cut unnecessary context in the prompt, and whether to switch some requests to a cheaper model.

Please provide access to the current settings of the bot (prompt plus examples of the knowledge base) — I will analyze and outline specific recommendations directly in the correspondence.

Budget: 250 PLN Deadline: 2 days

Hello, Polina! I am specifically dealing with RAG bots. You have a classic setup. The bot responds strictly from the knowledge base, and in case of a gap, it directs to the curator. Straight to the point, regarding your two questions:

1. Cost of responses. It is determined by the model and the volume of context that the bot pulls in each request. Most often, the overpayment is not in the platform itself, but in the fact that too much unnecessary text from the database goes into the model. This can be fixed by properly slicing the database and retrieving (sending less but more accurately), plus selecting the model for the task. I will look at your settings and tell you where it can be cut without losing quality.

2. Format of the knowledge base. Here it’s not either-or; short frequent questions work better in pairs of "question-answer," while long regulations are better as documents with a proper breakdown into fragments. At the same time, we will set the relevance threshold and the system prompt so that in the absence of an answer, the bot honestly directs to the curator instead of making things up.

Let’s take a look at your current configuration (settings, database, model, cost of responses), and I will provide a brief written analysis with specific corrections and answers to your questions. The timeframe is 1-2 days. To assess accurately, I need to clarify:
- What model is the bot currently using and what is the size of the knowledge base?
- Is there access to the dialogue logs and expense statistics in Suvvy?

Budget: 625 PLN Deadline: 3 days

It is correct that you have already launched a bot in Telegram based on the studio's knowledge and want it to respond only with the information found, and in the absence of it, direct to the curator, because this directly reduces the risk of "hallucinations" and maintains quality for the models. After the consultation, you will finalize the optimal settings and format of the knowledge base for your current content structure to reduce the cost of responses without losing accuracy, and you will establish a clear procedure for uploading materials for maximum relevance. Considering that you already have a bot on the platform and it "generally works correctly," I will analyze the current parameters of pricing/generation and the structure of the knowledge base according to the actual scenarios of the studio. I conduct audits of chatbots with restrictions on knowledge sources and escalation rules to the curator, so that responses remain strictly within the content framework, and computational costs are controlled. I will conduct a consultation in 8 working hours. How do you currently define "suitable information" in the knowledge base at the logic level (confidence threshold/key search/duration of retention) and what exactly does the bot transmit to the model when the curator is triggered, text or a link to the curator's entry? After your responses in DMs, I will prepare specific adjustments to the settings and formatting of the knowledge base for your case and help agree on where to reduce the response cost.

Portfolio - Freelancehunt

Budget: 200 PLN Deadline: 1 day

Good day. I understand AI bots for cost optimization on API—I launch and support bots using RAG and LLM myself.
Consultation format:
1. Audit of the current configuration of Suvvy.ai (settings, knowledge base, model)
2. Analysis of response costs—where you are overpaying, how to optimize
3. Recommendations for the knowledge base: which format is better (Q&A vs documents), how to structure content so that the bot does not "fantasize"
What you will receive:
A clear plan: what to change in the settings right now to reduce costs
A knowledge base scheme tailored to your scenarios (webcam studio—a specific niche)
An alternative: whether to switch to another platform (or stay with Suvvy.ai)
Format: audio call 1-1.5 hours + a short written conclusion
Price: 2,000 UAH / $50.
Portfolio: @MRAi_Demo_Bot—AI bot in production, own RAG, token optimization. Expand

Budget: 500 PLN Deadline: 3 days

Hello. I have about 4 years of experience in AI. I have completed over 200 AI projects. I can create a custom bot for the studio based on any AI, or continue working on a constructor. However, I would recommend looking into custom development, as it will be easier to scale and integrate later on.

If needed, we can discuss any questions further.

Budget: 90 PLN Deadline: 1 day

Hi there,

I have solid experience with AI chatbots, Telegram integrations, RAG/knowledge base systems, and OpenAI optimization.

✔ Audit of your current Suvvy.ai setup and architecture
✔ Analysis of token usage and response costs with recommendations for optimization
✔ Review of prompts, context handling, and model selection
✔ Recommendations on the best knowledge base structure (Q&A vs documents vs hybrid approach)
✔ Ensuring the bot only uses information from the knowledge base and redirects users to a curator when information is missing

I have worked on AI assistants where accuracy, cost efficiency, and preventing hallucinations were critical requirements.

I would be happy to review your current implementation and provide practical recommendations to improve both performance and operating costs.

Thanks 🤍 Expand

Budget: 400 PLN Deadline: 2 days

Hello. I have extensive experience in building AI automations. I can help conduct an audit and optimization; also, as an option, we can consider other ways of working with the bot, including the use of local LLMs.

Budget: 100 PLN Deadline: 1 day

Hello!

I have experience with AI bots and LLM APIs (OpenAI, Claude, Groq) — I can provide a consultation and answer your questions.

Regarding your tasks:

1. Cost of the bot's work — I can look at the settings and suggest whether it's possible to optimize tokens and reduce the cost of responses.

2. Knowledge base — I will advise on how to structure it correctly: what is better in a question-answer format and what as documents.

Format: online consultation via Telegram or Zoom, 1-1.5 hours.

Cost: $30-50 per consultation. Expand

Budget: 100 PLN Deadline: 1 day

Good day, Polina.

Your bot on Suvvy.ai is already working, but the key question is whether you are using the correct model and context format for each type of request, as this directly affects the cost of each response. During the audit, you will receive an analysis of the current settings: whether the LLM model is appropriately selected for your volumes, whether unnecessary context is being passed in the prompt, and where tokens are being spent that could be saved.

We will separately analyze the structure of the knowledge base. The "question-answer" parallels work more effectively for frequent typical requests of models — the bot takes a ready-made answer without unnecessary context expansion. Documents, on the other hand, are advisable to keep for complex procedural instructions where fragment search is needed. Your bot will find relevant information on the first try, and search costs will remain minimal.

If Suvvy.ai does not provide the necessary flexibility in managing costs, we will consider an alternative — a custom RAG architecture with pgvector and local embedding models (Ollama bge-m3), where you control every search parameter and only pay for the infrastructure. The consultation will include log analysis, specific recommendations on data formats, and, if necessary, a cost comparison of the current solution with the alternative.

150–200 PLN; ≈2 days
Quick analysis of current settings and answers to key questions.
- Cost analysis of responses and recommendations for reduction
- Advice on data upload format (question-answer vs documents)
- Answers to minor clarifying questions about functionality

250–400 PLN; ≈3 days
Complete comprehensive analysis with the development of a detailed knowledge base structure.
- In-depth log audit and precise cost calculation
- Step-by-step instructions for structuring the knowledge base for maximum accuracy
- Bot configuration to minimize "hallucinations" and unnecessary costs
- Consultation session covering all technical nuances

500–700 PLN; ≈5 days
Full audit and development of the architecture for transitioning to a professional RAG solution.
- All advantages of the Standard Package
- Comparative analysis of suvvy.ai against custom architecture (OpenAI + pgvector)
- Calculation of economic benefits when transitioning to your own infrastructure
- Prototype of the knowledge base structure for migration to the new platform Expand

Budget: 100 PLN Deadline: 2 days

Hello, Polina! I will answer both questions directly:

1. The cost of responses on suvvy.ai can almost always be reduced: usually, the budget is consumed by the long history of dialogue and overly large fragments of the knowledge base. I will look at your settings and show you where the money is going and what to change. We will compare it with the option of having your own bot without the platform's markup.

2. Knowledge base format: for strict scenarios (the bot should not give advice on its own), the Q&A format with short, clear entries works better + a strict system prompt with an escalation rule to the curator. I will show you using your examples how to reformat the base.

Format: a call lasting 60-90 minutes or a detailed text analysis - whichever is more convenient. After the consultation, there will be a checklist of actions.

I am building my first reviews on the platform, so the price is symbolic. I am ready to start today.

Budget: 160 PLN Deadline: 1 day

Good day, I have over 5 years of experience in developing AI agents and Telegram bots with RAG architecture - exactly such systems where the bot pulls answers from a knowledge base rather than generating them on its own. I can review your settings on suvvy.ai, evaluate the structure of the knowledge base and the cost of tokens - and tell you where optimization is possible (whether to stay on the platform or switch to a custom solution). There are also clear rules regarding the formatting of the knowledge base (Q&A vs documents) that significantly affect the accuracy and cost of responses. How would you prefer to conduct the consultation - via text or video call? I look forward to your message in private!

Budget: 8000 PLN Deadline: 2 days

I already have a practically ready solution for bots with a knowledge base, which can be quickly adapted to your scenario and discussed right here and now.
As for the format, I would suggest a short audit of the settings of suvvy.ai, checking the logic of the knowledge base, the cost of responses, and the risk that the bot will start giving its own advice.
Regarding the timeline - 1-2 days, the cost of consultation and audit - 8000 UAH.

Look, there’s a nuance here - for your task, it’s important not just to upload documents, but to divide knowledge into stable instructions, Q&A responses, and cases where the model should direct to a curator.
This way, you can reduce the cost of requests and eliminate unnecessary creativity from the AI, because the bot should be a reference guide, not an independent consultant.
It seems, based on my feelings, that you already have a decent base to start with, but without reviewing the settings, I wouldn’t pretend to see through walls - I’m not an X-ray with a subscription after all.

Let me clarify two points.
What is the current approximate consumption for 100 or 1000 responses?
Do you have examples of dialogues where the bot responded expensively, inaccurately, or delved into its own recommendations?

Relevant examples of our experience.
https://business.ingello.com/vorfahr - AI automation and agent logic for business processes.
https://business.ingello.com/fractal - approach to building AI agents and managed scenarios.

The main landing page of Ingello for the marketplace - https://systems-fl.ingello.com. Expand

Budget: 500 PLN Deadline: 7 days

Good afternoon, Polina! Is it possible to arrange it so that you only pay for the VPS (about 300 UAH/month) + my work?

Budget: 100 PLN Deadline: 1 day

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

Budget: 250 PLN Deadline: 1 day

Hello.

I develop bots for Telegram using NodeJS and various builders. I have experience with AI. Write to me, and we will discuss.

Budget: 130 PLN Deadline: 2 days

Hello, Polina! The task is completely clear. The situation where a bot on the builder (suvvy.ai) starts consuming too much budget or providing irrelevant answers due to a chaotic knowledge base is a classic problem when independently assembling AI assistants.

I specialize in developing AI bots in Python and integrating the OpenAI API, including setting up strict system prompts (hallucination guard) and optimizing context. I would be happy to conduct an audit for you and answer all your questions.

Work format: I will thoroughly analyze your settings, after which we will have a call (or I can create a structured text document — whichever is more convenient for you) and discuss the key points:

1. Cost optimization: I will calculate your current expenses and show you where you are overpaying. We will discuss how to set the context size (chunk size) and token limits on the platform to reduce response costs. I will also provide a comparison: how much it costs on the builder and how much it would cost with your own code (Python/SQLite), where you only pay the pure cost in OpenAI without platform markups.
2. Knowledge base architecture: We will clearly analyze when the Q&A format (question-answer) is effective and when whole documents are needed. I will advise you on how to properly format text so that the platform's search algorithm uses fewer tokens and finds the necessary instructions for webcam studio operators more accurately.
3. Instructions and curation: We will review the prompt to ensure the model strictly follows the rule of "not giving outside advice" and seamlessly transfers the dialogue to the curator in the absence of information.

I am gaining my first reviews on the platform, so I will offer you a very detailed analysis at a very comfortable price.

Clarifying question: please let me know which specific OpenAI model you are currently using in the bot settings (for example, gpt-4o, gpt-4o-mini, gpt-3.5) and whether you have the long dialogue history saving feature enabled? (These are key factors that usually "consume" the budget). Expand

Budget: 100 PLN Deadline: 1 day

Hello.
I have extensive experience in developing Telegram bots.
Feel free to contact me privately, we will discuss the details and get to work.

Budget: 150 PLN Deadline: 3 days

Hello!

I have reviewed your task. It's an excellent and clear case for implementing a classic RAG (Retrieval-Augmented Generation) architecture. I can develop such an AI assistant for your studio that will strictly follow the instructions and work only with your knowledge base.

Budget: 305 PLN Deadline: 3 days

I see you already have a working bot on suvvy.ai for the knowledge base of the webcam studio, just set it up without a specialist — understandable situation. I will conduct an audit: I will look at the current settings, calculate where the cost of responses comes from and where it can be reduced, and provide clear recommendations on the format of the knowledge base — when it's better to use "question-answer" and when a document, so that the bot accurately finds what is needed and doesn't waste extra tokens. Additionally, I will answer any further questions about the functions. The format can be a call or messaging, whichever is more convenient. I am ready to complete this in about 3 days.

Budget: 100 PLN Deadline: 1 day

Hello. I propose to start with a detailed audit of your current solution on suvvy.ai, analyzing the effectiveness of the architecture, query parameters to the model, and the current structure of the knowledge base. My expertise will allow me to identify optimal approaches to structuring the knowledge base to enhance the relevance of responses and minimize tokens, as well as evaluate alternative platforms or hybrid solutions to reduce operational costs. I have significant experience in developing and optimizing RAG systems for chatbots, which includes my own developments on effective chunking and data indexing, as well as templates for cost assessment across various LLM providers. I suggest discussing all implementation details, final budget, and timelines in private messages.

Budget: 100 PLN Deadline: 1 day

Hello! I can conduct an audit of the bot on Suvvy.ai: check the settings, response costs, work with the knowledge base, and the transfer of inquiries to the curator.

I will advise on how to better structure the information and where costs can be reduced without losing quality. At the end, I will provide written recommendations.

Budget: 5000 PLN Deadline: 2 days

I need an outside perspective on the already functioning bot: is the way suvvy.ai is set up worth the money, and is the knowledge base structured correctly?

For the consultation, I suggest the following format: first, I will review your current settings for suvvy.ai, the model, chunk size, and the method of searching the database — this directly affects the token cost per response. Then we will analyze the knowledge base: for FAQ bots with a limited domain (studio rules, scripts), the Q&A format is usually more accurate and cheaper than uploading entire documents, but this depends on how suvvy handles retrieval. If the platform does not provide the necessary control over this — we will discuss what can realistically be gained from alternatives.

How many messages does the bot currently process per day, and what is the cost per response?