AI agent for research

Hello! Your project is the perfect candidate for Vibe Coding (development at maximum speed using AI code generation). I am familiar with the entire stack you described (FastAPI, PostgreSQL + pgvector, Docling, Qwen 2.5), understand how to connect it, and will deploy it in the shortest possible time.

Show original

Translate

Hello.

I carefully studied the technical specifications. The project looks like a full-fledged research RAG platform, not just another AI chat with a connected LLM. I especially liked that you immediately laid out the correct architecture: ingestion → extraction → retrieval → reranking → structured recommendation.

My profile is right at the intersection of AI, backend development, and data processing systems.

I work with Python 3.11+, FastAPI, PostgreSQL, Docker, vector databases, and modern LLM stacks. I have implemented semantic search systems, RAG pipelines, document processing, AI assistants with long-term memory, and automated data analysis services.

What I can implement within your project:
…
✔ Uploading and processing PDF documents;
✔ Integration of GROBID for extracting the structure of scientific publications;
✔ Integration of PubMed and Scopus;
✔ Storing documents, metadata, embeddings, and query history in PostgreSQL;
✔ Hybrid search BM25 + Embeddings;
✔ Cohere Rerank to improve output quality;
✔ RAG pipeline with controlled response generation;
✔ Entity and parameter extraction through LLM + Pydantic Schema;
✔ Providing recommendations with mandatory justification and links to sources;
✔ REST API for integrations;
✔ Web interface on Vue 3;
✔ Docker deployment and documentation.

The strength of my approach is not just to make the model respond, but to achieve reproducible results, where the user always sees the sources, found documents, and the logic behind the recommendations.

For such systems, the quality of the retrieval layer and data extraction from publications is especially important. Therefore, I build the architecture in such a way that the system relies on scientific materials and factual data, rather than on the model's assumptions.

I am ready to start immediately after agreeing on the details. I can also offer several architectural options for different budgets and data volumes.

I would be happy to discuss the project.

Show original

Translate

Good day! The project is very interesting and quite complex — it is no longer just a chatbot, but a specialized RAG system for working with scientific documents, structured data extraction, and generating recommendations based on sources. We can assist with the architecture and implementation of the MVP:
— uploading and parsing scientific documents
— splitting text into chunks and storing metadata
— embeddings + full-text search
— RAG logic for generating responses in Ukrainian/Russian
— extracting parameters of methodologies through LLM + pydantic schema
— PostgreSQL schema for documents, fragments, embeddings, entities, and query history
— FastAPI backend
— web interface
… — generating recommendations with links to sources.
For the stack, it makes sense to use Python / FastAPI / PostgreSQL / Docker, and for search — a hybrid approach: BM25 + embeddings + rerank. To start, we can implement the MVP:
uploading documents → searching for relevant fragments → RAG response → structured output of parameters → links to sources. After that, the system can be scaled to Scopus/PubMed API, QLoRA/Qwen, and deeper AI analytics. We are ready to discuss the scope of the database, document format, accuracy requirements, and the roadmap for the first stage.

Show original

Translate

Hello. According to the specifications — you have a complete description at the link to Google Docs, I need access to the document to provide specific deadlines and pricing. Please grant read access (either to the freelancehunt account or an email address) — after reading, I will return with a detailed estimate.

Regarding the type of task — research agent is currently one of the most interesting AI domains. Architecturally, it usually boils down to three layers: tool layer (web search via Tavily/SerpAPI/Brave, retrieval from your sources, browser control if necessary), reasoning layer (planner + act-observe-reflect cycle, most often on LangGraph or a custom state machine), and output layer (structured report with citations and confidence). The quality critically depends on two things: the depth of the fact-checking step (verifying the agent's claims before providing the final answer) and proper prompt engineering for reporting — without these, the agent either hallucinates or produces vague text.

Relevant experience — I have been writing LLM integrations (OpenAI/Anthropic) for three years, there is a production AI assistant on RAG/Qdrant in production, and my current side project uses parallel agent orchestration via MCP. RAG + agent loops + prompt engineering is my main focus.

I am waiting for access to the specifications, and after reading, I will respond with specific figures regarding volume, deadlines, and stack.

Show original

Translate

Hello. I have experience in developing AI systems with workflow logic (stages, statuses, process automation, API integrations). I can implement your scenario turnkey: publication → bids → approval → reservation → execution → feedback with AI logic and testing before launch. After reviewing the specifications, I will provide exact timelines and costs, but approximately an MVP of this level takes 3–6 weeks depending on the complexity of the AI part and integrations.

Show original

Translate

Hello, Asher!

I would approach it in stages: first, a basic system with document upload, search, and responses based on sources, then gradually improving the quality of extraction and "smart" recommendations. The main focus here will be on data quality and proper handling of scientific texts, not just on LLM.

To better assess the implementation and timelines, we need to clarify a few points: what volume and format of source documents are planned at the start (PDF, article databases, pharmacopoeias), and whether access to them is already available or if they will be integrated as development progresses. It is also important to understand how deeply parameters need to be extracted — is general identification of methods sufficient, or is strict structuring "as in a database" with fixed fields required.

I suggest we discuss the details.

Show original

Translate

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

Show original

Translate

Good day. I am ready to complete this project and have extensive experience in developing various applications.

Show original

Translate

Asher, hello
I looked at the technical specifications - this is already a full-fledged RAG system for research tasks, and honestly, it's nice to see such a well-thought-out project architecture. It's clear that you understand what you want to achieve.

I especially liked that you immediately separated document ingestion, entity extraction, retrieval/rerank, and a separate RAG logic on top of that. This approach significantly reduces the risk of the project turning into "LLM responds randomly."

I want to clarify one important point: do you need an MVP for internal use in the lab/team or an immediate system designed for commercial scaling and external users? The architecture depends heavily on this.

Show original

Translate

Good day. I am ready to implement it step by step, starting with the MVP with basic parsing, database, search, and RAG responses.

Show original

Translate

Good day: we often do similar projects for other niches, we have experience! 2500 dollars, 10 working days!

Show original

Translate

Here is the response:

---

I have studied the technical specifications. The task is clear — not just a chatbot, but a working tool for the laboratory with RAG architecture, semantic search, and structured answers with links to sources.

I am familiar with the stack: FastAPI + PostgreSQL, Vue.js 3, embeddings (multilingual-e5-large), BM25 + Cohere Rerank, integration with PubMed/Scopus/GROBID. I have worked with similar systems for searching scientific documents.

What I propose:
…
— PostgreSQL schema for storing documents, fragments, embeddings, entities, and query history
— Hybrid search: BM25 + vector with re-ranking
— RAG pipeline with justification of answers and links to sources
— REST API for document upload and answer generation
— Web interface on Vue.js 3
— Docker configuration for deployment on VPS/RunPod

**Timeline:** 5–6 weeks including testing
**Cost:** we can discuss in private, I am considering the project budget — I am ready to offer a reasonable price.

I am ready to start after agreeing on the details.

Show original

Translate

Hello! I have reviewed your project and am ready to start working. I can guarantee excellent results in a short time.

Show original

Translate

Hello
We are qualified to take on the extremely specialized task of creating a turnkey AI solution for chromatographic analysis, as described in your specifications. Our strategy will concentrate on providing a reliable, intelligent system that serves as a useful tool for research and laboratory duties rather than merely a chatbot.
Using the stack you have chosen, we will implement the essential features, such as:
• Document Ingestion & Parsing: Scientific articles, pharmacopoeias, and methodological materials can be efficiently uploaded and parsed using Docling and GROBID.
• Data Extraction & Search: Using BM25, multilingual-e5-large, and Cohere Rerank 3 in conjunction with structured data extraction from text for semantic and full-text search throughout the knowledge base.
• RAG Logic & Recommendation Engine: Developing an advanced Retrieval-Augmented Generation (RAG) system to create useful suggestions for chromatographic analysis conditions, together with explanations and references. For complicated matrices like plant extracts and biological fluids, this will include fine-tuning an LLM (Qwen2.5-7b with QLoRA) to guarantee high accuracy and relevance.
Web Interface & API: Developing a user-friendly web interface with Vue.js 3 and a robust FastAPI backend, supported by a PostgreSQL 15+ database for storing documents, metadata, text fragments, extracted entities, embeddings, and query history.
• Deployment: Utilizing Docker, VPS, and RunPod for scalable and efficient deployment.
Given the complexity of integrating multiple specialized tools, fine-tuning an LLM, and building a full-stack application with a strong emphasis on accuracy and justification, a realistic timeline for delivering this turnkey AI, including thorough testing, would be approximately 16 to 24 weeks.
… For a project of this nature, which involves advanced AI development, specialized data processing, and a full-stack implementation, the investment would typically range from $80,000 to $150,000+. This estimate accounts for the specialized expertise required for RAG architecture, LLM fine-tuning, document parsing, and secure, scalable deployment.
Prior to a more accurate assessment, there are two crucial questions:
1. How many and what kinds of scientific documents—such as papers and pharmacopoeias—will need to be first consumed and digested into the system?
2. Does the AI need to meet any particular performance standards or accuracy metrics for the chromatographic recommendations?

Hello! I can develop a ready-made AI solution for you as a full-fledged production system, not just a demo. With proper testing, clear project handover, and a structure that allows your team to maintain it smoothly without unnecessary stress.

I work with LLM products end-to-end: from prompt architecture and data processing logic to secure APIs, logging, and a stable user experience.

Even before working on the UI, I prefer to first establish the core behavior of the system through a simple spec-driven pipeline, so the model operates reliably and quality can be measured. My approach is to first build a robust core, then add guardrails, retries, and clear fallback scenarios to ensure the system doesn't break when real users behave unpredictably.

As soon as I gain access to the document you specified, I will break down all requirements into a checklist, and we can track progress on each feature through short demos.

One idea that will make your AI more reliable and understandable for users from day one is the **Evidence Panel**. It will calmly show what data the response is based on and why the system chose that particular action. For users, it won't appear overloaded, but it will help support, reduce disputes, and speed up testing, as the logic of the system's operation can be seen without guesswork.
…
https://storyai.cc
https://live.chatbullet.com

Show original

Translate

I studied the technical specifications, the task is clear and technically interesting. I am ready to discuss the details at a convenient time, feel free to write and we will discuss.

Show original

Translate

We already have a practically ready architecture and developments for a similar AI agent, so based on your specifications, the estimate is 16,000 USD and 60 calendar days including testing. We can discuss the details here; I'm available.

Look, there’s a nuance - I would do this not as a regular chatbot, but as a !!research service with verifiable sources!! because the specifications include pharmacopoeias, publications, chromatography methods, Scopus, PubMed, Docling, GROBID, and RAG.

It’s better to divide the AI part into database search, extraction of structured parameters, ranking of sources, generating responses with citations, and logging requests.

The implementation will include a web interface, API, PostgreSQL, document upload, embeddings, request history, Docker deployment, a test set, and quality checking of responses based on your examples.

The cost of paid sources and GPU should be calculated separately to avoid hiding variable expenses in development.
…
From you, we need examples of 20-30 real requests, a set of documents for the initial database, access to paid sources if Scopus is needed in the first version, and criteria for correct responses.

I would like to clarify two points.
- Is Scopus mandatory in the first version, or can we start with PubMed and the uploaded document database?
- Are QLoRA and QWEN2.5-7b definitely needed at the start, or is it more important to have a stable RAG with good source output first?

Relevant cases and experience:
- https://business.ingello.com/vorfahr - similar in that it has AI generation, data processing, and product logic in a working system.
- https://business.ingello.com/fractal - close in agent architecture and automation of complex processes.
- https://business.ingello.com/tts - indirectly related through an AI service for an applied task.

The main profile of Ingello Systems for FLH - https://systems-fl.ingello.com.

We can start with a short design phase if you want to reduce risks regarding sources and the quality of responses.

Show original

Translate