Flagship project
Edge Voice-Agent Control Platform
Console + control API + on-device speech service in one loop. Past half-year focus; functionally closed and entering launch.
Stack (aligned with the flagship)
Admin web
Vue console, multi-region business APIs, permission-aware UI.
Control API
Java / Spring Boot, MySQL, Redis; devices, agents, models, knowledge, team permissions.
Device service
Python / asyncio, WebSocket sessions; VAD→ASR→LLM→TTS or end-to-end speech.
Retrieval and multimodal
External knowledge/vector engines via adapters; ASR/TTS and vision-related capabilities.
Engineering notes
Lessons learned
Dual-mode config, session latency, permission layers.
LLM gotchas
Injection, hallucination, context, rate limits.
Performance tips
Latency and throughput practices.