How antoinepinto/docu-talk works
Project type and purpose
• Docu Talk is a full-stack, cloud-native web application that lets users upload documents and spin up AI chatbots against them.
• The front end is a React SPA; the back end is a FastAPI service. The system integrates with external services like MongoDB Atlas, Google Cloud Storage, a Gemini LLM, AWS SES, and OAuth providers (Google/Microsoft).
File structure
• Top-level:
• backend/ – FastAPI app (routers, config, credentials, docu_talk package, mailing, database code, machine learning agents)
• frontend/ – React + TypeScript + Vite (components for auth, chat, settings, dashboard)
• deploy/ – shell scripts for deploying front- and back-end containers
• Notice routers (auth, chatbots, settings), utilities (auth, parsing, file_io), and the docu_talk “agents” subpackage (chatbot and predictor) in the back end.
• Configuration files: Dockerfiles, pyproject.toml/requirements.txt, package.json, tsconfig.json, and environment examples.
README for additional insights
• Network architecture calls out Cloud Run hosting, a VPC/NAT setup to communicate safely with MongoDB Atlas, Cloud Storage for PDF files (signed URLs), and AWS SES for transactional email.
• Authentication via email/password + OAuth 2.0 (Google, Microsoft).