DocuChat AI

An intelligent document assistant that combines conversational AI with multi-modal document understanding. Users can chat naturally with uploaded PDFs, images of documents, and handwritten forms.

Comment

Inspiration

Demand for fact checking at scale by remote users with variety of devices. Advanced OCR requirements for certain documents in current workflow.

What it does

Project Concept: DocuChat AI An intelligent document assistant that combines conversational AI with multi-modal document understanding. Users can chat naturally, upload PDFs, images of documents, and handwritten forms to get insights, summaries, and answers - all secured behind OAuth2 authentication with a comfortable, responsive pastel UI.

How we built it

Key Technical Stack:

LLM: NVIDIA Llama-3.1-Nemotron-Nano-8B-v1 (reasoning model) Embeddings: NVIDIA NemoRetriever-300M-Embed-v1 (for RAG) Deployment: Amazon EKS or SageMaker Auth: OAuth2 (Google/GitHub) Frontend: React with responsive design, pastel color scheme Backend: FastAPI/Node.js with vector database for retrieval

Core Features:

🔐 OAuth2 authentication (Google/GitHub) 💬 Natural conversational AI using Llama-3.1-Nemotron-Nano-8B-v1 📄 PDF document upload and Q&A 🖼️ Image document processing with OCR ✍️ Handwritten form recognition 🎨 Responsive pastel UI optimized for extended use 🔍 RAG pipeline with NemoRetriever embeddings ☁️ Deployed on AWS EKS/SageMaker as NIM microservices

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for DocuChat AI

Share this project:

Updates