Voice Agents
Notes on voice-first AI from hackathon and exploration work.
Voice is a compelling interface. No typing, no screens, just conversation.
ElevenLabs was the most interesting voice stack I built with. These are notes from that work — including Jazz Scat, which placed 2nd at the ElevenLabs Worldwide Hackathon.
Built & Proven
Jazz Scat
2nd Place 🥈ElevenLabs Worldwide Hackathon • Singapore • December 2025
AI jam partner for vocal practice. Conversational AI DJ + real-time backing track generation + voice coaching. Built in 12 hours.
What I Explored
Conversational AI Agents
Voice-first agents that hold a natural conversation — the hard parts are latency, turn-taking, and interruption handling, not the speech itself.
Voice Synthesis
Natural text-to-speech for content — demos, explainers, and audio versions of written work.
Sound & Music Generation
AI-generated sound effects and backing tracks — the core of the Jazz Scat build.
Client Tools & Orchestration
Wiring a conversational agent to real actions — the same tool-use patterns as text agents, under harder latency constraints.
Tech Stack
Why This Is Archived
The Brewery's own voice UI was retired in August 2026 — the exploration ran its course, and the lab moved its focus to agentic engineering. These notes stay up because the lessons (latency budgets, turn-taking, tool orchestration under real-time constraints) transfer directly to any conversational agent work.