UPLOAD ANYDOCUMENT.ASK ANYTHING.
DocMind is a RAG-based document Q&A platform. Upload PDFs, DOCX, or text files — DocMind chunks, embeds, and indexes your content into ChromaDB. Ask questions in natural language and get cited answers grounded in your source material. Zero hallucination tolerance, evaluated with RAGAS.
What attention mechanism does the Transformer use and how many heads?
The Transformer uses Multi-Head Attention with 8 parallel heads1. Each head operates on a d_k dimension of 64, with a model dimension d_model of 5122
Built on production-grade infrastructure
From upload to insight.
A three-stage RAG pipeline engineered for grounded, traceable answers. Every step is observable, every chunk is cited.
Upload
Drag any PDF, DOCX, or text file. Async ingestion via Celery — your dashboard stays responsive.
Index
Documents are chunked semantically, embedded with Gemini, and stored in your isolated ChromaDB namespace.
Chat
Multi-query expansion + cross-encoder reranking deliver cited answers streamed token-by-token.
Built for grounded answers.
Every feature engineered with one goal: zero hallucinations. Cite the source or refuse to answer.
Token-by-token streaming.
Server-Sent Events deliver each token the moment Gemini generates it. No waiting for full responses — feel the model think.
Citation-first.
Every claim traces back to a chunk. If the document doesn't answer it, neither does the model.
Any document.
PDFs with tables, DOCX with formatting, raw text. We handle the parsing chaos so you don't have to.
Multi-tenant ready.
Per-user ChromaDB namespaces. Your documents never touch another tenant's vector space.
Run it anywhere.
Docker Compose ships everything: FastAPI, Celery, Redis, ChromaDB, Postgres, and the Next.js client. One command, full stack.
See it in action.
Sample document: 'Attention Is All You Need' — the original Transformer paper. Pick a question to watch the streaming response.
Pick a sample question below to start
Simple plans. No surprises.
Start free. Upgrade when you ship. Cancel any time.