How varunkarnati/vois_ai_assignment works
Project type and purpose
• This is a full-stack, voice-driven ordering system for restaurants.
• Users speak into a browser UI, the speech is transcribed, sent to an LLM for dialog management, tools apply business logic (CRUD on orders), and results are spoken back.
File structure
• Top level:
• public/ – static frontend (index.html + JS/CSS)
• api/index.py – serverless FastAPI entrypoint (Vercel)
• app/main.py – FastAPI app definition, WebSocket routes
• app/utils/ – core modules:
– stt.py & tts.py for speech I/O
– llm.py for LangChain orchestration
– tools.py for order-management logic & file_db
– file_db.py for JSON read/write
– graph_definition.py for dialogue flow definitions
• orders_db.json + order_histories/ – JSON files acting as simple persistent store
• requirements.txt, vercel.json, .github – config and CI
README for additional insights
• Backend: FastAPI + WebSockets
• LLM: LangChain + Groq API
• TTS: ElevenLabs
• Database: local JSON
• Deployment: Vercel serverless, static hosting for public/
Detail
• Order_histories folder as an archive sink.
• Configuration inputs (.env keys for GROQ_API_KEY, ELEVENLABS_API_KEY).