What I Build
High-demand AI capabilities deployed on the stack that fits your infrastructure.
Enterprise RAG
Connect LLMs to your private data with vector search. Built for accuracy and citation.
Autonomous Agents
Systems that reason, plan, and execute complex workflows to automate processes.
Realtime Voice AI
Sub-300ms latency voice interfaces with human-like turn-taking.
LLM Ops & Eval
Observability, cost tracking, and guardrails to run AI safely in production.
I ship on LangChain, LlamaIndex, and Vercel AI SDK — or build custom runtimes when your use case demands it.
How I Work
Discovery
We define scope, constraints, and success metrics together.
Build
Iterative development with weekly demos and feedback loops.
Ship
Production deployment with documentation and support handoff.
Engineering Philosophy
I build production AI systems—the kind enterprises can trust, operate, and evolve. I help teams move from AI excitement to AI reliability: shipping platforms that combine RAG knowledge bases, agentic workflows, real-time voice, and deep integrations with Google Workspace and enterprise data—without sacrificing security boundaries, observability, or performance.
Catalyst is my flagship platform: a multi-tenant AI runtime with multi-provider routing, persistent memory, and tooling built for real operations. If you need AI that survives the real world—not just a demo—I can help you design it, ship it, and harden it.
Technical Expertise
Catalyst AI
Multi-Tenant Runtime.
A production-grade AI platform built for operations, not just demos. Featuring hard tenant isolation, custom LLM routing, and a real-time voice layer clocking under 300ms latency.
Featured Work
π.Law
EnterpriseLegal case management with automated document analysis and vector retrieval.
The Per4ex.org Show
LiveA live talk show where I moderate conversations between AI guests. Real-time via Ably.
Silicon Smackdown
Voice AIReal-time AI talk show with full-duplex voice debates. 20+ personalities powered by Gemini Live API.
Ready to Ship?
Tell me about your project and let's figure out the best path forward.
Typically respond within 24 hours