AI agent for searching and analyzing a dataset of documents in the decisions register

Budget: 27000 UAH Deadline: 45 days

Evaluation of the first working stage - 220,000 UAH, 45 working days. This is not the cost of the entire large product forever, but of a normal working MVP - data research, indexing of the test sample, semantic search, Claude API, classification of results, statistics, and a basic web interface.

If the registry has a stable API or OpenData, we can keep it simple and gather the solution quite pragmatically. I would build the system so that the model does not invent conclusions but works only through the found sources - first search and filters, then relevance checking, then classification of the result, a brief summary, and a link to the original source. The key risk is the quality and completeness of the data in the registry =/

To start, we need examples of 20-30 typical situations, the desired category of documents for the first sample, rules for determining the result, and an available method for obtaining data from the registry.

I would like to clarify 2 things to avoid shooting at fog with a cannon - does the registry have an official API/OpenData for mass document retrieval, or do we need to create a separate data collection module? What levels of authorities and what statuses of results should be included in the first version?

Similar examples in terms of operational logic
> https://business.ingello.com/vorfahr - AI automation and data processing for an applied business process
> https://business.ingello.com/fractal - agent logic, processing complex scenarios, and building a system around AI
> https://business.ingello.com/platforma - as an indirect example of a complex corporate platform with roles, data, and analytics

Main page of Ingello for FLH - https://systems-fl.ingello.com/ua Expand

Budget: 10000 UAH Deadline: 1 day

Hello!

We can develop an AI solution for you to search and analyze documents for this task.

1. Is there already access to the registry API or do you plan to work through open data?
2. What is a higher priority for the MVP: RAG search, Claude analytics, or a basic interface?

—
About us

We are dZENcode – a full-cycle IT development company: from design and programming to integrations and post-release support of digital solutions. We create projects from scratch, as well as connect to existing solutions that need refinement, development, or technical support.

You can find detailed information about our services and rates on our official website: Freelancehunt

Take a look – after that we can discuss the details and agree on the next step.

⚠️ After clarifying all the details, we will determine the scope of work, the suitable format of cooperation: task-based, outsourcing, or outstaffing – and the final cost.

With us, projects are guaranteed to reach release:
• 10+ years providing IT services;
• 90+ in-house specialists;
• 250+ public reviews since 2015;
• We support products under SLA after launch;
• We work under NDA and contract with the company! Expand

Budget: 27000 UAH Deadline: 20 days

Hello. I have experience in developing AI agents. I am ready for collaboration. Feel free to contact me.

Budget: 8000 UAH Deadline: 7 days

Good day! I am building RAG agents on Claude in production — semantic search across the registry, analysis, and reporting with references to primary sources is exactly my daily work. To accurately assess: what kind of registry it is (Unified Register of Court Decisions or another) and by what criterion we count the result — satisfied/refusal or a finer gradation? This affects the accuracy of classification. I suggest starting with an MVP on your real sample, then we can scale up the filters and statistics. I can show a live example of the agent right in the chat.

Budget: 27000 UAH Deadline: 21 days

Good day! This is our profile — AI agents over document arrays (RAG: semantic search and analysis). Regarding your task:

— semantic search instead of manual keyword selection: the agent finds relevant solutions based on the content and circumstances of the case, not just word matches;
— relevance assessment + a brief explanation of why the document fits the specific case;
— statistics of results based on circumstances (not "by eye") and a separate analysis of mass practice against the practice of higher/regulatory authorities.

What you will receive: a tool that reduces hours of manual review to minutes — it selects precedents, ranks them by relevance, and provides a summary.

To specify the exact timeline and price, a few questions: what is the volume of the registry (how many documents) and how is access to them — is there an API/export or do we parse an open registry? Where should the agent operate — a separate web interface or integration into your process? Approximately ~27,000 UAH (within your budget), ~2–4 weeks; I will confirm after the technical specifications. Work through Safe, in stages — no risk for you.

Examples of work: Freelancehunt · https://guardlabs.online Expand

Budget: 9000 UAH Deadline: 7 days

Good day! We are building and maintaining agents specifically on Claude, so your description reads like our daily work — RAG over an external source, not just a chat over a model.

As I see the architecture: free text query → semantic search over a vector database (Qdrant, it's lightweight and self-hosted, with no monthly fee like Pinecone) → Claude processes the top documents, discards irrelevant ones by coincidence, classifies the result (satisfied / partially / rejected) and compiles a report with links to the original sources. We calculate statistics on the sample based on the structured fields that the agent extracted from the documents.

One question to accurately assess the timelines: does the registry provide data through an open API or do we need to pull an OpenData dump and vectorize it ourselves? This affects the integration module.

I suggest starting with a working MVP — a complete cycle "query → search → analysis → report" on a limited sample, so you can see the quality on your real cases. Then we can scale up the filters by authority level and period and the complete statistical module. I am currently gathering initial feedback on Freelancehunt, so I am offering a starting price for the MVP stage.

I am ready to show a live demo of our agent right in the chat, so you can evaluate the approach to the solution.

Budget: 27000 UAH Deadline: 1 day

Hello, please contact me
_______________________________________________

Budget: 14000 UAH Deadline: 10 days

Hello!

My name is Oleksiy, and I represent the NC-1 engineering team. We have reviewed your project to create an AI agent based on Claude for working with document registries. We understand that for the stable operation of the RAG architecture (Retrieval-Augmented Generation), it is not enough to simply write code — a reliable DevOps/MLOps infrastructure is needed to ensure fast vector search, data security, and scalability.

Why our experience in DevOps/MLOps is critically important for your MVP:

Infrastructure for RAG: We specialize in deploying and configuring vector databases (Qdrant, Milvus, Pinecone). We will ensure minimal latency between the user query and Claude's response, which is key for quality UX.

Automation (CI/CD): We will build pipelines that automatically update the vectorization of new documents from your registry. Using GitLab CI or Jenkins, we guarantee that the AI always works with up-to-date data without your manual involvement.

Scalability and containerization: We use Kubernetes and Docker to ensure your system can easily handle load. We will configure the environment to be stable when scaling the database to millions of documents.

Monitoring AI performance: With the ELK Stack, Prometheus, and Grafana, we will monitor not only the state of the servers but also the performance of the AI: token usage, request processing speed, and document relevance quality.

Security and privacy: Working with document registries requires high security standards. We implement HashiCorp Vault for secret management and configure isolated network contours to keep your data completely secure.

How we will help your MVP:

Orchestration: Rapid deployment of the entire stack (vector database + API orchestrator) in the chosen cloud (AWS/GCP/Azure).

MLOps Lifecycle: Setting up automatic processes for re-indexing documents when the registry is updated.

Cost optimization: Configuring the infrastructure to achieve a balance between speed and cost of requests to the Claude API.

We have extensive experience managing complex infrastructures and building pipelines for high-load systems. We are ready to take on the technical part so that your team can focus solely on the logic of the AI agent and the quality of the prompts.

Are you ready for a brief technical call to discuss the deployment and scaling strategy of your database?

Best regards,

Oleksiy, NC-1 engineering team Expand

Budget: 10000 UAH Deadline: 14 days

Hello! I understand the problem — manual searching through the document registry is a pain. It can be solved through a RAG agent with hybrid retrieval.

Specific architecture:

Indexing the registry — chunking documents + embedding (text-embedding-3-small) → Qdrant
Hybrid search — semantic + BM25 keyword, re-ranker for accuracy
LangGraph agent — receives the request → searches for precedents → synthesizes the answer with references to sources
Interface — Telegram bot or web chat (FastAPI + Streamlit)

If needed, you can look at a project from the portfolio similar to yours - AI-radar — a RAG bot with hybrid retrieval, you can test it now (there's a link).

Budget: 27000 UAH Deadline: 35 days

We already have a practically ready similar solution for searching through a large array of documents, a vector database, analytics through AI, and a report with references; it can be quickly adapted and launched for your registry.
We are in touch and can discuss the details here on the marketplace and move to the initial design phase.

For the MVP, I would estimate around 35 working days and approximately 180,000 UAH, if the registry has a stable API or suitable OpenData.
If the data needs to be extracted in a more complex way, the estimate may change after researching the source - here the nuance is that the quality of access to the data will be the main risk.

Technically, I see the implementation as follows - a separate module for document collection, text normalization, embeddings, Qdrant or Milvus, hybrid search, Claude for result classification and brief summaries, then statistics and a web interface with filters, graphs, and a table of sources.
It is important not to let the model simply respond off the top of its head - it should only work on the found documents and show the original sources.

Questions for an accurate estimate:
- Which specific registry needs to be connected and does it have an API or OpenData?
- Which categories of documents are we taking for the first MVP and what is the expected volume - 10,000, 100,000, or more?

Relevant cases from Ingello:
- https://business.ingello.com/vorfahr - AI and SaaS logic with process automation
- https://business.ingello.com/fractal - agent approaches and complex automation based on AI
- https://business.ingello.com/platforma - corporate platform with roles, data, and analytics

Main landing page for FLH:
https://systems-fl.ingello.com/ua

Overall, this is a normal task for phased launch - first data research and a prototype on a sample, then a full index and precise classification testing.
Without checking the quality on real cases, such an agent will speak well but may make mistakes - and we need a working tool here, not an orator. Expand

Budget: 22000 UAH Deadline: 14 days

I understood the task: an AI agent on Claude is needed, which, based on a description of the situation in plain language, independently finds relevant documents in the decision register, analyzes their essence, calculates statistics of results under similar circumstances, and provides a ready analytical report with references to primary sources. This means eliminating the manual selection of keywords, manual reading of dozens of documents, and eye assessment of practices.

The architecture is exactly as you described; it is RAG. Documents from the register (via their API or by exporting open data) are broken into fragments, processed through embeddings, and stored in a vector database (Qdrant, or Pinecone or Milvus, we will choose based on volume and budget). Upon request, the system performs a semantic search based on content, retrieves the top relevant documents, and Claude classifies them, extracts the essence, and compiles a report with statistics and direct links to sources.

Two points determine the quality here. First, to ensure the agent does not fabricate: the response is built only on the found documents, each conclusion with a reference to a specific decision, nothing from itself. Second, to ensure the statistics are honest: the classification of results must be done according to fixed rules and show the sample on which it is calculated; otherwise, the numbers look good but are unreliable. We will separately distinguish mass practice at a lower level and precedent level at a higher authority, as you requested.

This is my main area: I created a voice AI assistant with RAG search through a knowledge base using Qdrant, where the LLM also responds only based on real data and does not fabricate. So both semantic search across a large array and grounded analysis through Claude are working tasks for me.

Please advise on the main point for evaluation: does the register provide documents through an official API, or do we need to set up our own vector database on exported open data? And what is the approximate volume of the document array?

Budget: 9000 UAH Deadline: 7 days

Good day! It seems that the key challenge here is not in finding documents, but in building a quality RAG solution with correct case classification and analytics without AI hallucinations.

I have experience in building AI solutions based on Claude, ChatGPT, Make.com, and integrating external data sources into a single automated process.

I can also implement an MVP with a web interface or a bot, where the user will receive a ready analytical report with links to sources, statistics, and brief conclusions for each case.

Have you already decided on the data source: the registry API or are you planning to work through OpenData exports?

Budget: 20000 UAH Deadline: 14 days

Good day. I have a lot of experience in creating AI agents. I have completed similar tasks before, so it shouldn't take much time. We can connect in private messages to discuss my experience in more detail, and I can also send you my CV.

Budget: 10000 UAH Deadline: 1 day

Hello! I am interested in your project. I have experience with RAG architecture and integrating the Claude API, so I fully understand the technical requirements for building the search and analysis system you outlined.

Why you should choose me for implementing the MVP:

Technical expertise: I understand how to set up the vectorization pipeline (Pinecone/Milvus/Qdrant) and ensure high-quality semantic search.

Working with Claude API: I have experience writing system prompts for Claude, which minimizes hallucinations when extracting facts and generating reports.

Systematic approach: I am ready to take on the full development cycle—from data research and sample preparation to creating a user-friendly interface.

I work as a sole proprietor (group 3), so if needed, I can provide all closing documents for your reporting.

I would be happy to discuss the implementation details, especially the nuances of integration with registries. Please write to me, as messages are the most convenient and quickest communication channel for me. Expand

Budget: 1000 UAH Deadline: 1 day

Good day, Oleg.

In brief:

Your AI agent based on Claude will independently search for documents in the registry, analyze them, and generate analytical reports with references to primary sources. The system will use semantic search in a vector database, allowing it to find relevant cases based on a free-form description of the situation, even without an exact match of keywords. For each document, the AI will determine the outcome and generate a summary, while the statistical module will display the distribution of practices on the dashboard. All processing will take place on your infrastructure, ensuring confidentiality.

In more detail:

In the first stage: the solution is based on a proven production RAG pipeline in Go with PostgreSQL/pgvector and local embeddings Ollama bge-m3. Using a local vectorization model instead of external APIs significantly reduces costs and delays when processing arrays of 10,000+ documents. Integration of Claude via OpenRouter using clear system prompts ensures fact extraction and classification of decisions (approved/partially approved/denied) without hallucinations. The user interacts with the system through a minimalist web application or a user-friendly Telegram bot for testing, where they can see analytics and the distribution of cases by levels of authorities.

Then, a hybrid search (BM25 + Vector Search) can be implemented for instant retrieval of documents by specific case numbers or rare terms, which is critical for legal accuracy. At this stage, a real-time automatic monitoring module for the registry will also be added, which will notify about the appearance of new documents based on selected criteria and update the statistics of precedents without manual re-launching of the analysis.

Later, a precedent expert system can be created, which will accumulate and take into account the experience and edits of your specialists. This will significantly reduce costs and speed up the work on the solution by decreasing the number of calls to large language models, as well as further enhance the accuracy of analytics. Expand

Budget: 22000 UAH Deadline: 25 days

Good day.
Your project is exactly the stack I work with. I have real experience building a RAG system with semantic search, vector database, and integrating Claude API into production.
Here’s how I see the implementation:
FastAPI backend for integration with the registry. ChromaDB for vector storage and semantic search. Two-level filtering — Haiku quickly filters out irrelevant content, Sonnet deeply analyzes the selected items. Structured JSON output for accurate classification without hallucinations — satisfied / partially satisfied / denied. A final report with a table, statistics, and links to primary sources.
Before starting, it’s important to clarify — which registry, is there an official API or open data, and the approximate volume of documents. The website with examples of work — neo-space.site
I am ready to discuss the details.

Budget: 27000 UAH Deadline: 17 days

Most RAG-based solutions do not take into account the deep specifics of legal documents. Your task requires not just textual representation, but systematic analysis of precedents across multiple dimensions. For a statistically correct report, it is important to configure Claude to distinguish the context of differentiation from the arguments of decisions. Experience shows that the most common mistake occurs at the stage of filtering by levels of authorities — 40% of relevant data is lost due to incorrect classification. It is necessary to ensure automatic cross-checking of case circumstances. Have you already selected a specific category of documents for testing, or do you need to form a sample from all available data?

Budget: 27000 UAH Deadline: 25 days

Good day. The project is interesting, but to properly assess the MVP, I would like to clarify a few points:
Do you have access to the registry API, or do we need to work through open data?
What specific category of documents is needed to start?
What volume is required for the MVP: a test sample or a large dataset right away?
Is it sufficient at the initial stage of analysis to focus on one category of cases?
Is a web interface needed right away, or would a working MVP with basic search, statistics, and reporting be sufficient at first?
Are there examples of 5–10 documents that you consider relevant to better fine-tune the selection logic?
Am I correct in understanding that the main task at the initial stage is to find similar documents, determine the outcome, show statistics, and provide a brief AI conclusion with references to sources?

Budget: 27000 UAH Deadline: 14 days

Hello, I will develop RAG for your database, write if the project is real.

Budget: 27000 UAH Deadline: 30 days

Good day. I am ready to implement. I have already worked on similar tasks in projects:
https://o-keto.com/ — AI-nutritionist with RAG database based on Google Docs
https://gloap.net/ — AI-job search and AI-resume selection with RAG database based on database content
https://gloap.net/ — AI-recruiter: selection of sailors through searches on sailor websites, communication with sailors in messengers.

Budget: 20000 UAH Deadline: 10 days

Good day.

I can help with the implementation of an MVP for an AI agent for searching and analyzing documents in the decisions registry.

I see the solution not just as "a chat with Claude," but as a RAG system: retrieving documents from the registry or open data, preparing texts, semantic search through embeddings/vector database, analyzing the found documents through Claude, and generating a structured report with references to sources.

For the first stage, I would suggest creating an MVP on a limited category of documents or a test sample. Within the MVP, we can implement:

* searching for relevant documents based on a natural language description of the situation;
* filtering by basic parameters;
* analyzing each found document through Claude;
* determining the result: satisfied / partially satisfied / denied / undetermined;
* a brief summary for each document;
* a results table with references to original sources;
* basic statistics on relevant documents;
* a final analytical conclusion for the user.

An important point: for a quality result, it is necessary to first check the availability of registry data/API/OpenData and the accuracy of classification on the test sample. After that, the system can be scaled to a larger array of documents.

I have experience in building AI automations, n8n workflows, API integrations, working with OpenAI/Claude, structured JSON output, Google Sheets/CRM, and automated analytical reports.

I am ready to start with the MVP to quickly verify the technical implementation and the quality of analysis before full scaling. Expand

Budget: 27000 UAH Deadline: 30 days

Good day!

I am ready to take on the development of such an AI agent. It is important to create not just a chat with Claude, but a proper RAG pipeline: document collection, indexing, semantic search, relevance filtering, result classification, and the formation of an analytical report with references to sources.

I would suggest starting with an MVP / proof-of-concept on a limited sample of documents to quickly check the quality of search and analysis.

I see the work plan as follows:

1. Analysis of the data source and document format.
2. Parsing and preparation of the test database.
3. Vectorization of documents and search through Qdrant / Pinecone.
4. Integration of Claude API for relevance analysis, results, and a brief summary.
5. Formation of statistics on cases.
6. A simple web interface or bot for testing.

Technically, I can implement it in Python / Node.js, Claude API, embeddings, Qdrant or Pinecone, PostgreSQL, FastAPI / Express, React / Next.js.

Approximately:
— PoC on a small sample: 20,000–30,000 UAH
— MVP with a basic interface and statistics: 50,000–60,000 UAH

The timeline for the MVP is about 3–5 weeks, depending on the availability of the registry/API and the quality of the source data.

I will be able to provide a final estimate after clarifying the data source, the volume of documents, and the format of the required interface. Expand