Inspiration
Reading long research papers is slow, and finding one specific answer means skimming dozens of pages. I wanted a tool that lets you simply ask your PDFs a question.
What it does
ResearchMind AI is a RAG-based research assistant. Upload a PDF, ask questions in plain language, and get answers grounded in the document's own content.
How I built it
- PyMuPDF extracts text from the PDF
- Sentence Transformers turn text chunks into embeddings
- FAISS retrieves the most relevant chunks for each question
- Llama 3.2 (via Ollama) generates the answer from those chunks
- Streamlit provides the interface
The code is split into four modules: app.py, pdf_utils.py, vector_store.py, and llm.py.
Challenges
Limited RAM meant I had to switch to the smaller llama3.2:1b model and tune chunking so answers stayed accurate.
What I learned
How retrieval quality drives answer quality, and how to design an AI product around real hardware constraints.
What's next for ResearchMind AI
Multi-document search, source citations, and chat history.
Built With
- faiss
- llama-3.2
- ollama
- pymupdf
- python
- rag
- sentence-transformers
- streamlit
