How zdrlfx/fullstop works
Project type and purpose
• This is a full‐stack voice & chat assistant (“JUNU”) with a React.js frontend and a FastAPI backend.
• Purpose: allow citizens to interact via text or voice to ask government-process questions. Backend handles chat (OpenAI/GPT), speech-to-text (Azure STT), and text-to-speech (Azure TTS).
File structure
• Top-level:
• frontend/ – React UI code (pages, components, build configs)
• server/ + main.py – FastAPI endpoints and logic (inference, model wrappers, speech processing, document indexing)
• Data/ – markdown docs used by the backend indexer
• config files: package.json, requirements.txt, .env expectations
• Architectural cues: separation of concerns (UI vs. API), modular backend (chat, STT, TTS, docs), no persistent database (in-memory or on-the-fly indexing).
Key insights from the README
• Backend exposes three REST endpoints: /chat, /stt, /tts
• Relies on Azure SDKs and OpenAI API
• Frontend built with Vite + Tailwind, deployed as a static bundle served by backend or CDN.
Detail level
• Backend modules (inference.py, model.py, speechprocessing.py, index_docs.py) as sub‐components
• Where Data/*.md is loaded and served to the Chat Service for context
• Lack of a persistent database (in-memory only) if relevant