Intelligent Document Analysis
DocLens uses local embeddings and state-of-the-art AI to let you have natural conversations with any PDF — instantly, privately, for free.
Why DocLens
Embeddings are generated locally on your machine. Your document content never leaves for vectorization.
Powered by Groq's inference engine running Llama 3.1. Responses in under 2 seconds, every time.
HuggingFace local embeddings paired with Groq's generous free API tier — no credit card needed.
FAISS vector store with top-k semantic search finds the most relevant chunks for every question.
Natural, conversational Q&A with full session history. Ask follow-ups just like you're talking to a person.
Clone, install requirements, add your API key and run. Up and running in under 5 minutes.
Under the hood
DocLens extracts raw text from your PDF using pypdf, handling multi-page documents with ease.
Text is split into 1,000-character overlapping chunks, then embedded locally with all-MiniLM-L6-v2 — no data sent externally.
Your question is embedded and compared against the FAISS index. The top 3 most relevant chunks are retrieved.
Context + question are sent to Groq's Llama 3.1 8B Instant model, which generates a grounded, accurate answer.