How joychudasama/chat-txt works
Project type and purpose
• ChatTXT is a full‑stack web application for interactive AI chat over user‑uploaded TXT documents.
• Frontend is built in React/TypeScript with Tailwind and custom UI components; backend is a Python FastAPI service using Langchain, Ollama LLM, and Firebase/Firestore.
Main architectural components
a. Frontend
• React SPA (App.tsx, components folder)
• UI library: Shadcn/ui + Tailwind CSS
• WebSocketContext for live chat or REST calls for upload/auth/session
b. Backend
• FastAPI application (backend/app/main.py)
• API router grouping v1 endpoints (auth, chat, file_upload, session)
• Layered packages:
– core: app setup, configuration
– db: Firestore connector
– models: Pydantic request/response schemas
– prompts: prebuilt query templates (qa_prompts)
– services: business logic (file_service, rag_service, session_service, vector_store_service)
c. Data stores & external services
• Firestore: user/session data, chat history, metadata, possibly embedded vectors
• Ollama: local LLM for text completion/inference
• (Implicit) vector store via Langchain wrapper on Firestore
d. Client-browser interactions
• HTTP(S) REST calls for file upload, session management, standard chat
• WebSocket (optional) for real‑time chat UI updates
Relationships & data flows
• User → Frontend UI →
• POST /file_upload → backend.file_service → vector_store_service → Langchain embedding → Firestore
• POST /chat → backend.rag_service → vector_store_service (Firestone or metadata index) → Ollama LLM → response → Firestore (session & history) → frontend
• GET/POST /session → session_service → Firestore
• Frontend subscribes to WebSocketContext for streaming chat messages
• All components interact over well‑defined API endpoints
Architectural patterns & design principles
• Layered architecture: API routers (controllers) → services (business logic) → data layer (db/vector store)
• Dependency inversion: core/config drives service instantiation
• Separation of concerns: prompts templates are decoupled from service logic
• Extensibility: new endpoints or services can be plugged into the same router/service structure
Technologies & frameworks callouts
• Frontend: React, TypeScript, Vite, Tailwind, Shadcn/ui, Zod for validation
• Backend: FastAPI, Pydantic, Langchain, Ollama API, Firebase Admin SDK for Firestore
• Dev tooling: ESLint, TypeScript configs, Python requirements.txt, .env examples
• Notes on protocols (HTTP, WebSocket)
• Environment/config layer feeding into core/app_setup