Back to Projects
RAG
Completed
Document Q&A System
An intelligent PDF assistant that delivers fast, context-aware answers with source references.
2026• Team: Personal Project
About this Project
Document Q&A is an intelligent document analysis system that lets you upload PDFs and ask natural language questions about them. Get instant, context-grounded answers with source citations—no more manual document scanning.
Tech Stack
PyMuPDF (fitz)
SentenceTransformer + FAISS
Ollama (llama3)
Streamlit
Tools Used
VS Code
Git
Streamlit
Key Features
Core Features
- ▸Natural Language Q&A: Ask questions in plain English, get instant answers
- ▸Source Attribution: Every answer includes citations from your document
- ▸Interactive UI: Clean, organized interface with sidebar controls and chat history
- ▸Local Processing: All processing happens locally—your documents never leave your machine
- ▸Error Handling: Full error handling with Ollama status checks and graceful failures
- ▸Debug Visibility: Console logs track all pipeline stages with [DEBUG] prefixes
Architecture & Design
- ▸**[PDF EXTRACTION] → Extract text, clean, create chunks (500 words, 50-word overlap)
- ▸**[EMBEDDING] → Convert chunks to 384-dimensional vectors (SentenceTransformer)
- ▸**[INDEXING] → Build searchable FAISS index (L2-distance similarity)
- ▸**[RETRIEVAL] → Find top-K similar chunks for query (default K=3)
- ▸**[GENERATION] → Inject context into prompt, query LLM (Ollama llama3)
- ▸**User Gets Grounded Answer + Sources
Highlights
PyMuPDF (fitz)
SentenceTransformer + FAISS
Ollama (llama3)
Streamlit