VoxArchitect — Voice-Native Intelligence

# VoxArchitect Voice-native enterprise architecture and change intelligence built for the AssemblyAI Voice Agent Hackathon. VoxArchitect lets an architect speak a change request, inspects repository/system evidence, maps dependencies, reasons about architecture and security impact, drafts ADRs, and stages GitHub actions behind an explicit human approval gate. ## Architecture - **Frontend:** React + Vite, browser microphone capture, 24 kHz PCM AudioWorklet, AssemblyAI Voice Agent WebSocket, live transcript, evidence/graph view. - **Backend:** FastAPI REST API for temporary AssemblyAI tokens, tool dispatch, orchestration, GitHub integration, approval-gated actions and knowledge/evidence storage. - **Agents:** Intent Router, Repo Scout, Dependency Mapper, Architecture Analyst, Trust Reviewer and Delivery Agent coordinated by a deterministic orchestrator. - **LLM:** Optional AssemblyAI LLM Gateway (OpenAI-compatible). The app still runs in demo/fallback mode without credentials. - **Hosting:** Docker Compose. Nginx serves the built frontend and reverse-proxies `/api` to FastAPI.