"Roxi AI team on AssemblyAI - Voice Agent Hackathon Hackathon"
Team Idea
Roxi AI is a real-time multi-agent meeting co-pilot that records audio of meetings to provide live strategic and technical insights. Going beyond post-meeting transcription, Roxi analyzes audio instantly using a Rust/Tauri core with OS-level loopback. A Node.js sidecar handles seamless failover of STT providers via a ring-buffer to avoid data loss. For reduced LLM latency and cost, a Rust ChunkGraph orchestrator analyzes utterance complexity within milliseconds. Chit-chat is ignored, and parallel AI agents—such as Risk Detectors and Technical Analysts—are launched instantly only on dense topics, showing live insights. All data is locally saved in SQLite database, analyzed post-meeting using an Adaptive-RAG chatbot. Roxi AI is built after overcoming challenging macOS audio routing and IPC framing issues. Future plans include advanced speaker diarization, proactively taking action during meetings, and full offline local LLM capabilities.
Team Members
.png&w=256&q=75)
Manthan Mittal
Manthan26
