AI assistant (company knowledge base)

Budget: 4000 UAH Deadline: 4 days

Hello!

I can quickly conduct an audit of your prototype and find out why the context is not reaching the model. I have commercial experience working with Python, AI models, REST API, Linux, and integrating LLM solutions. In my work, I have dealt with integrating ONNX models, AI pipelines, STT, TTS, and optimizing models for real products.

I would start with the audit: checking context transmission, API calls, document indexing, embeddings, and retrieval. If the issue can be fixed, I will do it. If the current architecture is not feasible to maintain, I will propose a new one with a detailed explanation.

For implementation, I would use Python, FastAPI, OpenAI or Claude API, pgvector or Qdrant, Docker, and a web interface with authentication. The architecture can be prepared for further scaling right away.

I am ready to start immediately after receiving access.

Thank you! Expand

Budget: 8000 UAH Deadline: 5 days

Good day.

First, I will conduct an audit of your prototype. Usually, the problem lies in the fact that the text from the database does not reach the neural network due to an error in the connection settings (API). If there is no point in salvaging the old code, I will not write the web interface, authorization, and chat history from scratch. Developing such a project from scratch costs much more than the stated 4000 hryvnias.

I propose a workable alternative to your request. I will deploy a ready-made platform for working with language models (for example, Dify) on your server. You will immediately receive a ready web chat with password access for the sales department. The neural network (we will connect OpenAI API) will be strictly limited: it will only search for answers in your uploaded texts and will honestly respond "I don't know" if there is no information. You will update the knowledge base yourself through a simple control panel by dragging new PDFs or documents there without the involvement of a programmer. The entire history of dialogues will also be available for quality control. Regarding monthly expenses: about 15 dollars for the server and approximately 10-20 dollars for tokens.

Where are your regulations and service descriptions currently physically located (just files on a computer, Google Drive, or a database in Notion), so I understand the format for their initial upload?

Budget: 27000 UAH Deadline: 7 days

Hello. I have experience in developing AI automation. I am ready to collaborate.

Budget: 4000 UAH Deadline: 7 days

Good day! I have worked on exactly such RAG assistants — a web interface with authorization, a company document database, and responses strictly from sources. Your symptom "information from the database does not reach the model" almost always means that the documents have not been indexed into the vector database or the found fragments are not being inserted into the prompt before the query — rarely the API itself. I will look at the code and access, and in half a day I will give you an exact diagnosis and what is more beneficial: to fix it or to rebuild it. What is the prototype made on — LangChain/LlamaIndex or a custom solution, and what storage is used for embeddings?

Budget: 27000 UAH Deadline: 3 days

The cost of the first stage is 32,000 UAH, duration - 3 working days. This is an audit of the existing prototype with checks on the API, context transfer to the model, document indexing, vector database, authorization, and current code. 4,000 UAH, in my opinion, does not even cover a normal technical audit of such a task, let alone a full service =/

After the audit, we will provide one of two routes - to fix the existing one or to rebuild from scratch. If rebuilding, the working minimum for a RAG assistant with web access, authorization, query history, database updates without a developer, HTTPS, and readiness for future connection of an analytical repository I would estimate separately, approximately 15-30 working days and from 180,000 UAH.

Stack - Python or Node.js for the server, PostgreSQL plus pgvector or Qdrant for vector search, OpenAI or Claude via API, a separate document processor, web interface for managers, role-based authorization, query log. For responses based on documents, it is necessary to add a link to the source or at least the title of the found document; otherwise, managers will not be able to properly control quality.

Estimated ownership - hosting and database 20-80 USD per month, tokens 20-150 USD per month at the start. We will calculate more accurately after the number of managers, average document lengths, and the number of queries per day. The model should not invent facts, so we make responses only with RAG and a scenario when there is no data in the database.

Two clarifications
> Approximately how many documents or pages are in the knowledge base, and in what formats are they currently collected - PDF, Word, Google Docs, spreadsheets, website?
> Does the current prototype already have a vector database, or are the documents just uploaded to the server without proper content search?

Similar cases from Ingello
> https://business.ingello.com/vorfahr - AI automation with applied business logic, close to an internal assistant for the sales team.
> https://business.ingello.com/fractal - agent and system architecture, useful for tasks where chat needs to be part of a managed process.
> https://business.ingello.com/forma-crm - corporate logic, roles, processes, quality control of team work.

Freelancehunt page of Ingello - https://systems-fl.ingello.com/ua

!!The optimal first step is an audit, not a promise of the entire product within a budget of 4,000 UAH.!! It is better to first find where the knowledge transfer to the model is breaking down, and then decide whether to fix the existing one or build anew. Expand

Budget: 4000 UAH Deadline: 7 days

Hello, I worked on an AI assistant for a law firm — a RAG system based on 200+ documents, Pinecone vector database, Claude API, ~$45/month for tokens with 1000 requests.

Regarding your project — can you clarify which vector database is currently used in the prototype and how the documents are transmitted to the model? This will help quickly identify the issue with the API during the audit.

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

Budget: 4000 UAH Deadline: 2 days

Hello. I looked at your project — a classic RAG for an internal knowledge base. First, I will conduct an audit of the existing prototype: the most common issue is that the embeddings are not updated after loading documents or the retrieval chain is incorrectly configured. If the code is live — I will fix it; if there are deeper architectural problems — I will rebuild it on LangChain + ChromaDB/Qdrant, with a web interface on Streamlit or FastAPI + a simple front end. Regarding LLM: Claude Haiku is optimal in terms of price/quality for this case, approximately $15-30/month per sales department. The monthly cost of ownership (tokens + hosting) is within $30-50 depending on intensity. I have created similar solutions for internal documentation. I am ready to start with the audit, the full cycle — up to 2 weeks.

Budget: 4000 UAH Deadline: 2 days

Hello! It is correct that the first step is an audit, not blind rework. Your prototype likely does not pass the found document snippets into the model request itself: the data is loaded, but either retrieval is not connected to the LLM call, or it returns empty, so the model responds with general knowledge instead of your database. I will look at the code and access and honestly say whether it’s faster to fix or rebuild the core. Often, assembling a clean RAG pipeline is cheaper than fixing someone else's.

The key in the result is for the bot to respond strictly from your documents and honestly say "this is not in the database" when there is no answer, rather than making things up. This is precisely the grounding I emphasize.

Stack: vector database (pgvector or Chroma) + embeddings, LLM via API, web interface with authorization via link, file uploads through the interface for updating the database without a developer, query history. Regarding the model: for English documents and accuracy, I would recommend Claude, with the option to drop simple queries to a cheaper model for token savings. I will immediately make the architecture modular so that later you can connect an analytical repository, as you plan.

I am ready to analyze for free in private why your prototype does not pull data into the model. I would be happy to collaborate.

Budget: 6000 UAH Deadline: 5 days

RAG with strict grounding is my specialization: responses only from documents, honestly reporting when not in the database, sources are traceable.

The first step is an audit of your prototype (judging by the description — a break in the context transfer to the model). Next: pgvector + embeddings, web interface with authorization, file upload/update without a developer, query history. I will ensure compatibility with your analytical storage.

Within your budget - basic version 6,000 UAH, 5 days: audit + working RAG on your documentation, web access with authorization. Full version - 12,000 UAH, 8 days: + file upload through the interface without a developer, query history, readiness for connection to the analytical storage.

Ownership: hosting ~$10/month + tokens ~$15–30/month. Portfolio: in the profile and on karasov.co.

Budget: 4000 UAH Deadline: 1 day

Hello, I have relevant development experience.
Message me privately, we will discuss the details.
I will be happy to help you!

Budget: 5000 UAH Deadline: 5 days

Good day, I am very interested in working with you. I offer quality execution of the tasks set and promptness. Let's discuss the details in a private chat and start our collaboration.

Budget: 26990 UAH Deadline: 29 days

Good day!
My name is Valentin, and I represent Arctic Web Agency. We are a team that specializes in creating modern and effective solutions for businesses. I can provide examples of our similar work in personal messages. We are ready to take your project to work!

Sincerely,
Arctic Web Team
Freelancehunt

Budget: 4000 UAH Deadline: 5 days

Hello! I have carefully reviewed your specifications. Development and troubleshooting of RAG systems (Retrieval-Augmented Generation) is my main profile. The issue with your prototype ("information not being passed to the model") is most likely due to incorrect payload formation for the API or the vector search returning empty chunks (fragments of text) due to errors in the document splitting algorithm.

My work plan (Audit and Rebuild):

Audit: I will access the server, check the logic of embeddings and queries. If the architecture is hopeless, I will reasonably rebuild the pipeline on modern frameworks (LangChain or LlamaIndex).

Strict RAG without hallucinations: I will configure the system prompt (temperature = 0) and the fallback logic: if there is no relevant chunk in the vector database, the model will strictly respond: "I don't have this information in the provided documents," instead of fabricating facts.

Interface and Admin Panel: To fit within the budget and for a quick release, I will deploy a web interface based on Streamlit. There will be a chat window for managers (with basic password authorization), a query history, and a separate tab for the admin where new PDF/Docx files can be simply dragged and dropped for automatic database updates.

Scalability (for DWH): The architecture will be built modularly (Agentic approach). In the future, we will be able to add a new "tool" to the assistant so that it can perform SQL queries to your analytical storage in addition to the vector database.

Proposed stack: Python, LangChain/LlamaIndex, ChromaDB/Qdrant (vector database), Streamlit (UI), FastAPI (if a bare API is needed).

Main project risk: Pricing tables. Standard parsers often "break" the structure of tables during text chunking, causing the model to confuse prices. I will use specialized parsers (for example, Unstructured) that maintain relationships in tables. Expand

Budget: 4000 UAH Deadline: 3 days

Hello.

For the first stage, I can take on the audit and bring the prototype to a working state if the problem is indeed in transferring context from the knowledge base to the LLM query.

In my own document processing system, I have already applied a similar outline: documents go through text extraction, normalization, chunking, knowledge storage separate from raw files, relevant part search, and response formation linked to the source. Separately, I have a FastAPI RAG prototype for AI content: ingestion, chunking, retrieval with citations, route/service/repository structure, and tested core logic.

For your prototype, I will first check:
- whether embeddings are created after document uploads
- whether there is proper chunking and metadata for sources
- whether vector search returns relevant fragments
- whether these fragments actually make it into the prompt/context before calling Claude/OpenAI
- whether the history of queries and pipeline errors is stored

The stack for a stable version: Python/FastAPI, PostgreSQL + pgvector or Qdrant, OpenAI/Claude API, Docker, HTTPS, a simple web interface with authorization, a log of requests and responses with links to the source.

Estimated ownership cost to start:
- VPS/hosting: approximately $10–30/month
- tokens: approximately $10–70/month for a small sales team, I will calculate more accurately after the number of documents, managers, and queries per day

My rate is 4000 UAH / 3 days — for the first stage: audit, localization of the problem, and, if the architecture allows, launching a working prototype. If after the audit it makes more sense to rebuild the system from scratch, I will provide a separate estimate for the complete MVP.

We can discuss the nuances in private messages. Expand

Budget: 4000 UAH Deadline: 3 days

Hello! RAG systems (LLM strictly based on documents, without fabrications) are my specialization.

Let's start with the audit, as you suggested. The symptom "data from the database does not reach the model" is almost always one of the typical RAG mistakes: embeddings are not pulled into the prompt, empty output from vector search, incorrect chunking of documents, or context is not passed in the API call. In 1-2 days, I will find the reason and provide a reasoned recommendation: fix the existing one or rebuild it (with justification).

Next — a working system:
• Uploading all documentation through the interface (database updates without a developer)
• Strict RAG: response only from documents, "not in the database — honestly says"
• Web access via a link with authorization
• Scenarios: questions about the company / inserted client question → ready answer / regulation search
• Query history for quality control
• Architecture for future connection to the analytical repository

Stack: Python (FastAPI), Claude/OpenAI API, RAG pipeline with vector DB (pgvector) + embeddings, PostgreSQL, web UI, HTTPS, hosting on your server or in the cloud.

Regarding LLM: for QA on documents, I recommend economical models (Claude Haiku / GPT-4o-mini) — the quality for this task is excellent, and tokens are significantly cheaper.

I will show examples of RAG solutions and a live demo in private. Question: what is the approximate volume of documentation (pages/files) and how many managers will be using it?
https://horizonbyte-hub.vercel.app/uk Expand

Budget: 4000 UAH Deadline: 10 days

It is necessary to fix (or rebuild) the RAG assistant, which accurately answers managers' questions based solely on company documents.

The first step is an audit of the existing code: I will look at where the chain breaks between the vector database and the LLM call (most often this is either an incorrectly formed context in the prompt or embeddings not being indexed). If the prototype is really broken deeper — I will rebuild it on LangChain + pgvector or Chroma, embeddings through OpenAI text-embedding-3-small, LLM through GPT-4o-mini (cheaper than Claude Sonnet at the same level for FAQ scenarios). The web interface on FastAPI + minimal React: token-based authorization, file uploads through the UI, logging requests in PostgreSQL. The architecture immediately with abstraction for future connection of an analytical repository.

Estimated monthly cost: hosting ~300-500 UAH, tokens at 50 requests/day ~200-400 UAH.

What is the current vector database in the prototype and what is the backend written on, so I can assess the scope of the audit before starting?

Budget: 4000 UAH Deadline: 7 days

I will fix your prototype: first, a code audit, and within a day, I will provide an accurate diagnosis and conclusion on whether it is more beneficial to repair or rebuild from scratch. One nuance from your specifications that is easy to overlook: you require English language support, and when the manager inserts the client's questions as they are, while the documents are written in another language, standard embeddings miss the search. Therefore, I will implement multilingual embeddings and test the search in both languages before delivery. Responses will strictly be based on the found fragments with the source document's title, and if there are no responses in the database, the assistant will directly state that. This approach can be tested right now: find the bot auth_ai_radar_bot in Telegram, this is my RAG with hybrid search, it responds only based on the uploaded documents and honestly declines when there is no data. Stack: Python + FastAPI, PostgreSQL with pgvector or Qdrant, Claude API (I will also calculate an option on OpenAI), web interface with authorization, query history, database updates with files through the interface. Estimated monthly ownership for the sales department: $10-30 for tokens plus $10-20 for hosting, I will calculate more accurately based on your document volume. We will discuss the details in the chat.

Budget: 24000 UAH Deadline: 7 days

Hello! The task description is very well written. The issue with "information not being passed to the model" likely lies in the logic of the RAG pipeline: incorrect chunking of documents, a low cosine similarity threshold in vector search, or a failure at the stage of gathering the final context for the prompt.

I specialize in developing reliable web applications and API integrations and propose to implement your RAG service as a stable monolith on the Laravel 11-13 framework. This will avoid bloating the infrastructure (no complex connections of third-party Python scripts) and keep the entire project in one clean codebase.

Technical stack and architecture:
- Backend and AI pipeline: PHP 8.3-8.4 / Laravel 11-13. Complete logic for chunking, context preparation, and working with APIs (OpenAI / Anthropic).
- Vector storage: PostgreSQL + pgvector extension. Vectors (embeddings) will be stored directly in your database, ensuring maximum retrieval speed without third-party services.
- Interface and admin panel: Filament PHP. Covers all your requirements turnkey: ready-made manager authorization forms, clean chat interface, query history for quality control, and a convenient panel for uploading new PDF/Docx documents without involving a developer.

Ownership economics (for a department of 10 managers):
1. Hosting: Regular VPS (e.g., Hetzner, 4 GB RAM) - ~$10-$15/month.
2. Tokens: Using gpt-4o-mini or Claude 3.5 Haiku will cost about ~$10-$20 per month with active use, as basic models are currently very inexpensive.

Commercial terms:
The budget specified in the order (4,000 UAH) can only cover the initial audit of your current prototype and technical consultation. A full assembly of a stable RAG system, configuration of vector search, prompt engineering against hallucinations, and deployment will cost more.

- Actual turnkey development cost: 18,000 - 24,000 UAH (depending on the volume of legacy code).
- Timeline: 5-7 working days.

Design the architecture so that in the future you can easily connect the system to your analytical storage (BI) through Laravel services. I am ready to look at your access and current code in private messages. Expand

Budget: 27000 UAH Deadline: 20 days

Hello, Yuriy

We can create a RAG system for your business. Please specify the number of documents and their format that you currently have.

Budget: 4000 UAH Deadline: 3 days

Good day, I have extensive experience in developing agents and working with documents and RAG indexing. There are a couple of closed cases in my profile. We can discuss in more detail; I can complete it in a few days.

Budget: 4000 UAH Deadline: 5 days

Good day! I understood the task: the assistant must respond strictly according to your documentation (RAG) and not make things up, and currently, the prototype is not transmitting data from the database to the model — most likely, the break is exactly at the junction of "database → LLM" (the context is not reaching the API request or the vector database is empty/unassembled).

The first step is an audit: I will connect to the prototype and the server, look at how the pipeline is structured (where the data comes from and where it goes, what is used to create embeddings, how the request to the model is formed), and I will honestly say — whether to fix the existing one or it’s cheaper to rebuild.

If building from scratch: RAG on Claude/OpenAI — vector database (embeddings according to your documentation) + answer service strictly based on found fragments with an honest "I don't know" when there is no answer in the database; web access via a link with authorization; query history; database updates through the interface.

Please provide access to the existing prototype and server — in half a day, I will localize where exactly the transmission to the model is breaking.

Regarding the price: 4000 UAH realistically covers the audit + bringing the prototype to a working state. A complete outline (web UI with authorization, query history, self-updating database) is a larger scope, so I will name the final cost after the audit when the state of the code is clear. I will provide the token calculation based on your volume of documentation separately.

Budget: 4000 UAH Deadline: 5 days

Hello! I have reviewed the project and am ready to start working. I am confident you will be satisfied with the result.

Budget: 4000 UAH Deadline: 5 days

Recommended stack: OpenAI or Claude via API, RAG (embeddings + vector database), FastAPI, PostgreSQL + pgvector or Qdrant. Or Ollama and a 100% local model, among which there is a wide selection, depending on infrastructure and tasks. I will design the architecture with future integration to your data storage in mind.

Budget: 4000 UAH Deadline: 1 day

Good day!

I will start with an audit of your code. On the first day, we can discuss whether to fix the existing one or rebuild it. The problem may lie in the information search algorithm. As an alternative - my service momicro.com. This will meet your needs. ABSOLUTELY EVERYTHING can be configured, customized, and tailored specifically for you. All these integrations are free! I will adapt it to your sales process in a day.

GitHub: https://github.com/axbuglak

Sincerely,
Buglak Oleksiy

Budget: 4000 UAH Deadline: 5 days

Good day! We have experience in deploying RAG systems based on LLM. We implement this through a vector database and integration with your documentation for accurate answers. We will set up a data processing pipeline so that managers receive relevant information in real time. We are ready to start refining your prototype and bring it to a stable working condition.

Budget: 10000 UAH Deadline: 3 days

Hello!
We have experience in developing AI assistants and RAG systems based on OpenAI and Claude. We are ready to first conduct an audit of the current solution, find the cause of the incorrect API operation, and restore functionality, or, if more appropriate, build the system from scratch.
We implement a full-fledged knowledge base with RAG, a web interface with authorization, a query history, simple document updates without a developer, and we will also prepare the architecture for further integration with analytical data. We will help select the optimal LLM considering the quality of responses and the cost of use.
After familiarizing ourselves with the current project, we will prepare a detailed assessment of timelines, budget, and estimated monthly expenses for tokens and hosting. We would be happy to discuss the details of cooperation.

Budget: 4500 UAH Deadline: 1 day

Hello!

I am a Full-Stack Software Engineer with over 7 years of experience in developing websites, SaaS solutions, complex web platforms, and MVPs for startups - from idea and architecture to production and support.

I work not only as a developer but also with a focus on business logic, scalability, and long-term support of solutions. My portfolio includes examples of completed projects of varying complexity.

Technology stack:
PHP (Laravel, Symfony, Yii2),
Frontend: JavaScript (Vue.js, React.js), HTML5, CSS3,
Databases: MySQL, PostgreSQL.

Regarding timelines and costs - at this stage, it is difficult to accurately assess them without clarifying the requirements. To form a reasoned estimate, it is usually necessary to understand:
- project goals;
- key functionality;
- expected load and integrations;
- priorities (quick launch vs scaling).

I suggest starting with a brief clarification of the requirements or a call/email - after that, I will be able to provide a realistic estimate on timelines, budget, and implementation options.

I would be happy to discuss the details. Expand

Budget: 27000 UAH Deadline: 14 days

I can take on this project for a quick audit of the current prototype, fix the existing integration if it's reasonable to keep it, or quickly rebuild the service from scratch if that would be more reliable and faster. In the end, you will receive a RAG assistant that strictly responds based on the company's documents, does not fabricate facts, has a web interface with authorization for managers, a query history, and convenient knowledge base updates without a developer.

Proposed stack: OpenAI or Claude API, Python/FastAPI, PostgreSQL + pgvector or Qdrant, React/Next.js for the interface, a pipeline for loading and indexing documents, deployment on your server or in the cloud with HTTPS. The architecture can be designed from the start to allow future connection of the assistant to an analytical data warehouse.

I have experience with similar RAG solutions: internal knowledge bases, documentation search, FAQ assistants, systems for sales/support teams with responses based on sources.

Budget: 4000 UAH Deadline: 3 days

Good day!
I am finishing the project https://salesslon.com - it uses RAG for analyzing conversations between managers and clients. The model does not invent anything, it works exclusively with the texts of the conversations.

I will take a look at what is wrong with your project, why it is not working. If there are no critical issues, I will fix the error. If everything is bad, I will tell you how I would do it.

The cost of RAG is very low, it depends on the volume of data, there are even free models available.

Budget: 4000 UAH Deadline: 5 days

Good day!

I have worked on similar AI solutions where the main task was to create a unified knowledge system and automate data handling. From the description, I see that you already have a prototype, but the problem likely lies in integrating the RAG pipeline with the LLM or passing context through the API. I would start with a technical audit to determine whether it is more feasible to fix the current architecture or to rebuild it more quickly and reliably.

In my projects, I have built AI assistants, automated data processing through Make.com, integrated CRM, AI services, and APIs into unified business processes, and created AI consultants for lead qualification and personalized responses.

Estimated timelines:
• audit and restoration of the current solution — 2–4 days;
• if a complete rebuild is needed — 7–10 working days.

Cost:
• audit — from $300;
• full implementation — $1,200–1,800 (depending on the current state of the code and requirements).

Monthly costs:
• VPS hosting — about $15–40;
• OpenAI/Claude tokens — approximately $20–150+ depending on the number of managers and requests.

Please let me know which model is currently being used (OpenAI or Claude) and which vector database you have already connected? This will help quickly assess whether it is sufficient to fix the existing solution or if it is worth immediately proposing a more reliable architecture. Expand

Budget: 15000 UAH Deadline: 7 days

The main thing now is to find the reason why the prototype is not transmitting data from the knowledge base to the LLM. If the problem is only with the API or RAG pipeline, it can be fixed without a complete overhaul. However, if the architecture is poorly built, it would be more logical to rebuild it right away to avoid future scaling issues.

What needs to be done:
• conduct an audit of the current project and check the API integration;
• configure RAG so that the model responds only based on the documentation;
• implement loading and updating of the knowledge base without developer involvement;
• add a web interface with authorization and query history;
• prepare the system for further connection of other data sources.

It is also worth deciding right away which model to use (OpenAI or Claude), as this will affect the quality of responses and monthly token costs.

If you provide access to the current project, it will be possible to quickly understand whether it is sufficient to fix the existing solution or if it is more advantageous to build a new architecture. Expand

Budget: 4000 UAH Deadline: 3 days

RAG-assistant for sales EN: audit of your prototype → fix retrieval or reassembly (Python/FastAPI, pgvector, Claude/GPT). Web + auth, upload without a developer, query history, mode "insert client's question - get answer." ~3–5 weeks, ~90–150 hours; hosting+API ~$80–350/month. Send the code — I will conduct the audit in 1–3 days.

Budget: 4000 UAH Deadline: 1 day

Good day! 🧡
I am ready to start working and would be happy to help with minimal deadlines 😉
I have experience working on similar projects. 🌼 (I have reviews on this topic and work in my portfolio)
I will do it in the best way possible and will start now! 🚀🛸✈️

Feel free to write, and we will discuss. I respond instantly!

Budget: 4000 UAH Deadline: 1 day

Hello!

The most likely reason why the documents are not reaching the model is that retrieval finds fragments, but they are not substituted into the context before the API call — or the embeddings are simply not updated after loading new documents. A code and access audit will reveal the exact point of failure in half a day to a day.

Stage 1 — Audit. I will check API calls, indexing, vector database, and prompt assembly. Based on the results, I will provide a clear conclusion: whether to fix the existing setup or rebuild from scratch, with justification and cost/time estimates.
Stage 2 — Implementation based on the audit results. Stack: FastAPI, PostgreSQL + pgvector (or Qdrant), Claude API, web interface with authorization, query history, database updates without a developer.
I have experience with FastAPI and working with external APIs (including building a Telegram bot that monitors several marketplaces via API in real-time — similar logic for processing and delivering data).

Clarification: what is the current prototype written in (LangChain/LlamaIndex or a custom solution) and which vector database is already connected?

Feel free to reach out! Expand

Budget: 11111 UAH Deadline: 1 day

Hello! I have over 3 years of experience creating AI/Full-stack services. I have a ready, tested RAG template that meets 90% of your requirements.

My stack: FastAPI, LangGraph (agents and history), Qdrant (vector search with MMR), Flashrank (re-ranking against hallucinations), Redis, and Docker.

Implementation plan:

Audit (1 day): I will check your prototype. If the problem is only with the API — I will fix it. If the code is raw — I will reasonably deploy my ready backend, which will save time.

RAG and Control (3–4 days): I will set up strict system prompts (strict QA) — the model will respond strictly according to the documents. I will integrate Claude 3.5 Sonnet or GPT-4o mini.

Interface and Admin Panel (3 days): I will create a clean web interface with authorization, a section for uploading/deleting files by managers (without developer involvement), and logging chat history. The architecture will initially be ready for connecting an analytical database.

Timeline and cost:

Timeline: 5–7 days (most likely faster, but I am taking a buffer).

Cost of work: $400–$7000.

Estimated monthly expenses: Hosting (VPS) — $20–$30; Tokens (for the sales team) — $15–$40 depending on the load.

I am ready to show a demo of a similar system, write to discuss! Expand

Budget: 3800 UAH Deadline: 5 days

Hello! I am ready to start with a technical audit of the existing prototype - I will check the API, the transfer of document context to the model, indexing, and the reason why the database is not being pulled. Stack: Python, PostgreSQL + pgvector (or Qdrant) for vector search, RAG pipeline with OpenAI/Claude API, web interface with authorization for managers, database updates without a developer through file uploads. Responses will be strictly based on the uploaded documents, with a fair fallback if the information is not available. After the audit, I will provide a clear answer - to fix the existing one or to rebuild from scratch.

Budget: 5000 UAH Deadline: 5 days

Hello! This is exactly the task I'm working on — RAG integration, where a ready prototype needs to be "linked" to a knowledge base.

Recently, I delivered a FastAPI panel for managing accounts with background tasks and real-time status (7500 UAH) — a similar architecture: a service that processes data and responds to requests.

For your case, the standard stack: document loading → chunking → embeddings → vector database (Qdrant/Chroma/pgvector) → retrieval chain. If the prototype is already ready, the knowledge base integration will take 3–7 days depending on the document format and the prototype architecture.

The budget of 4000 UAH seems modest — if the volume is small, that's fine; if not, it’s worth discussing. In what format is the documentation stored (PDF, Word, Google Docs)? What stack is used in the prototype?

Budget: 4000 UAH Deadline: 1 day

Good day, Yuriy!
I am ready to take on your project. I am a certified Anthropic developer — Claude Certified Architect (Foundations), specializing specifically in production architecture with the Claude API, MCP, and RAG pipelines. Certificate verification: https://www.credly.com/badges/3df97b6b-e468-42b9-9bd6-c1f78aa309fc
A similar case. Recently, I implemented a similar project - an internal knowledge graph based on Neo4j + Claude through MCP, which answers employee queries strictly according to corporate documentation, with references to the source and without "fabricated" facts. This is almost exactly your task, only I additionally used a knowledge graph for more accurate searching based on the relationships between regulations.

Two clarifications to calculate more accurately:

How many documents/pages are approximately in the database and in what formats (PDF, Word, Google Docs, spreadsheets)?
In the current prototype, is there already a vector database, or are the documents simply uploaded to the server without content search?

I am ready to start with the audit immediately after receiving the code and access.

Budget: 5000 UAH Deadline: 10 days

Hello, I can implement this. I have built RAG systems based on n8n.

Google Drive

Everything except the LLM itself can be deployed on your server. This will be self-hosted n8n + supabase, as a vector storage.

The price will only be for tokens, and it all depends on the number of requests, the accuracy of responses, and the number of documents in the RAG system. How many documents will there be? Through which interface do you want to communicate with the system? And is there a system administrator who can deploy n8n and supabase?

Budget: 4000 UAH Deadline: 1 day

Good day, Yurii.

The most likely reason: the model receives a request, but no document fragments are found as context — retrieval is not connected to the API call. An audit will identify the break point and determine whether to fix or rebuild.

Your RAG pipeline — pgvector and bge-m3 for semantic search.
LLM: Claude Sonnet or GPT-4o-mini with token calculations tailored to your volumes.
The model will respond strictly based on context, without fabrications.
Web interface with authorization, query history, and database updates with files without a developer.
Architecture — with an extension to an analytical repository.

Similar project: RAG on Go + pgvector + OpenRouter.
Stack: Go, pgvector, bge-m3, Claude/OpenAI API, Docker, HTTPS.
I will prepare the monthly ownership cost after discussing the volumes.

Possible options:
- Recovery and launch
8000–12000 UAH
Recovery of the prototype's functionality and basic access to the knowledge base.
- A working RAG pipeline without model "hallucinations"
- Basic web interface for obtaining responses
- Connection of the chosen LLM API (OpenAI/Claude)
- Manual database update instructions

- Full functionality
15000–25000 UAH
A full-fledged corporate assistant with autonomous knowledge management (Recommended).
- Document upload interface without developer involvement
- Authorization system for sales department managers
- Query history for monitoring response quality
- Optimized architecture for future connection to the analytical repository
- Full calculation of token and hosting costs

- Advanced solution
30000–45000 UAH
An advanced AI tool with enhanced accuracy and support for complex scenarios.
- Improved document processing (OCR for scans, complex tables)
- Extended response generation functionality for various communication channels
- Priority support and cost optimization for tokens over the month
- Integration with cloud storage for automatic file synchronization
- Complete technical documentation package for system architecture Expand

Budget: 4000 UAH Deadline: 1 day

Hello!

If we are to work exclusively based on the established knowledge base, then the LLM is needed only for generating responses and understanding questions, while all vectorization (RAG) can be performed by a local model.

I created such an internal manager assistant for ddtuning.
Let me know, and I will show you how it works.

Regarding looking into why it is not working now, we can take a look.

Budget: 7000 UAH Deadline: 7 days

I have built similar RAG systems: FastAPI, vector database (Qdrant / pgvector), embeddings through Claude or OpenAI, web interface with authorization. The first step is to audit the prototype: I will find where the pipeline breaks between the database and the model, and we will decide what is faster — to fix it or to rebuild it.

Stack: FastAPI, pgvector/Qdrant, Claude API (recommended — more accurate on complex queries, transparent token pricing), simple web interface, Docker, HTTPS. Database updates — file uploads without a developer. The architecture is designed with a focus on a future analytical repository.

Approximately 35 hours of work. Ready to start after receiving access.

Write to me — we will discuss the details and get started.

Budget: 4000 UAH Deadline: 5 days

Hello!

I have experience in developing services based on LLM and RAG architecture. I suggest starting with an audit of the current solution: checking the integration with the API, the indexing and search pipeline, after which either restore functionality or reasonably propose a rebuild.

For implementation, I propose the stack: Python (FastAPI), OpenAI or Claude API, PostgreSQL + pgvector (or Qdrant), a modern web interface with authorization and the ability to independently update the knowledge base.

The estimated duration of the first stage (audit + launch of the working version) is 5–10 days. I will be able to provide the exact cost after reviewing the current code and architecture. I will also help estimate the monthly expenses for API and hosting depending on the volume of documents and the number of requests.