DocuMentor

Inspiration

We wanted to make it easier for users to interact with large collections of documents without manually searching through dozens of PDF files. DocuMentor was inspired by the need for an intelligent assistant that could understand multiple documents, retrieve relevant information, and provide grounded answers with transparent sources.

What it does

DocuMentor is a full-stack Agentic RAG application for multi-document PDF question answering and semantic search. Users can upload and query multiple PDFs through a conversational interface. The system retrieves relevant information, performs multi-step reasoning when required, and provides answers along with source citations for transparency.

How we built it

We built DocuMentor using React.js for the frontend and FastAPI for backend services. A LangGraph agentic pipeline coordinates planning, retrieval, and evaluation nodes to support multi-hop reasoning. MongoDB Atlas Vector Search stores and retrieves document embeddings, while all-MiniLM-L6-v2 generates semantic embeddings. Groq LLMs power the language reasoning and response generation across multiple rate-limited models.

Challenges we ran into

One of the main challenges was designing an agentic workflow capable of determining when additional retrieval or reasoning was necessary. We also had to handle large document collections efficiently, maintain accurate source attribution, and work around rate limits while using multiple LLM models.

Accomplishments that we're proud of

We successfully built an end-to-end Agentic RAG system capable of reasoning across multiple documents rather than relying on simple single-query retrieval. We are particularly proud of the plan-retrieve-evaluate workflow, streaming responses, semantic search, and transparent source-citation interface.

What we learned

We gained practical experience building production-oriented RAG and agentic systems, including vector search, embedding generation, document retrieval, prompt orchestration, and multi-step reasoning. We also learned how frontend streaming, backend APIs, and LLM pipelines can be integrated into a seamless user experience.

What's next for DocuMentor

We plan to improve retrieval accuracy with more advanced ranking and hybrid search techniques, introduce stronger document-level reasoning, and support additional file formats. We also want to improve agent evaluation, reduce latency, and expand the system's ability to handle complex multi-hop queries.

Built With

• agentic
• agentic-ai
• api
• fastapi
• generative
• langgraph
• llm
• mongodb
• natural-language-processing
• rag
• semantic
• vectordb