Dhruv Patel, AI and full stack engineer

A voice agent that needs the LLM less the more you use it

A local classifier picks the tool, and a language model steps in only on a miss. Its answer is cached for next time.

An event-driven voice agent that routes tools locally first and learns from its LLM fallback.

Role: Author
Status: Open source
Stack: Python, asyncio, Faster-Whisper, sentence-transformers, ChromaDB, Groq