I'm the founder of Rerato and maker of Trivana.
Most AI waits for a prompt. I'm interested in AI that can take responsibility for an experience: welcome the room, guide the flow, adapt in real time, remember with permission, and help a group reach an outcome.
Rerato is the platform I'm building around that idea. Trivana is the first live product—AI-hosted games and learning experiences made from source content.
Before this, I spent nine years building applied AI in regulated environments across cybersecurity, financial services, NLP, voice, graph ML, and MLOps. I still write code every day and like owning the whole loop—from model behavior to runtime to customer outcome.
- Voice and personality that stay coherent
- Memory with clear user permission
- Low-latency orchestration for live rooms
- Grounded generation and ruthless evaluation
- Systems that survive contact with real users
- voice-persona-engine — shape consistent AI personas
- agent-flow — orchestrate multi-agent workflows
- realtime-ai-serve — streaming inference primitives
- llm-eval-harness — catch output regressions before users do
- rerato-api-examples — safe integration patterns for Rerato
The product code behind Rerato and Trivana stays private; these repos are the technical pieces I can share.
If you're building AI that has to perform live—not just look good in a demo—I'd enjoy comparing notes.
Voice is the interface. Orchestration is the product.