DocuMind — Chat with Your PDFs

DocuMind turns any PDF into something you can actually talk to. Instead of scrolling through pages of dense text, you upload a document and ask it questions directly — DocuMind retrieves the most relevant sections and answers using Google's Gemma 4 model, with every response backed by real source citations so you can verify exactly where the answer came from. The problem is simple: research papers, contracts, and reports bury the information people need. Traditional search finds keywords, not answers. DocuMind reads for understanding and stays honest — if the answer isn't in the document, it says so instead of guessing. Key features: - Automatic PDF upload and text extraction - Natural language Q&A with grounded, accurate answers - Source citations on every response, showing the exact snippet used — proof the system isn't hallucinating - One-click summaries - Automatic action item extraction - "New Chat" reset for clean sessions Built with React and TypeScript on the frontend, Supabase for storage, database, and Edge Functions, and Google's Gemma 4 via the Gemini API powering every response — chosen specifically for this hackathon's Gemma challenge. Built solo, end-to-end, including real debugging of CORS issues, timeout handling, and a retrieval bug that initially caused missed titles/authors — fixed and now working with correct, cited answers. Intentionally scoped as a focused MVP: PDF-only for now, a capped context window, and single-user access, prioritizing a genuinely reliable core AI experience over spreading effort thin. The vision extends to researchers, teams, students, and professionals who need to trust an answer, not just get one — the citation system is the foundation for that trust.