Document Q&A system that turns PDFs into searchable knowledge bases. Upload a report, ask a question in plain English, get an answer with numbered source citations and relevance scores. The retrieval layer chunks documents into overlapping passages (600 chars with 100-char overlap to avoid splitting context at boundaries), builds a TF-IDF vector index, and ranks results by cosine similarity. The answer is extracted from the highest-scoring passage with cross-reference to the original text. Built the conversational web UI with a chat-style interface. Python + FastAPI backend, scikit-learn for vectorization, PyPDF for text extraction.