Principal AI Architect · Senior Director · Engineering Leader
Designing the architecture and leading the teams that ship it.
AI systems architect and engineering leader with 20+ years building real-time distributed systems. Currently leading a 15-person AI engineering team delivering production AI for the customer support domain — voice training systems, multi-agent workflows, and applied LLM architectures.
20+
Years in real-time systems
15
Engineers led currently
5+
Agent architectures shipped
GCP
Cloud-deployed infrastructure
01
Focus Areas
What I build
Three system types recur across the portfolio. Each is treated as a practical system with state management and measurable outputs — not a prompt demo.
01
Multi-Agent AI Systems
Structured agent workflows with orchestration, handoffs, and measurable evaluation — interview coaching, internal copilots, and assessment engines.
02
Real-Time Voice AI
Streaming speech-to-speech systems built on real-time architecture principles: low latency, session control, and structured feedback loops.
03
Applied AI for Products
RAG assistants, learning content pipelines, and automation agents — taken from idea to working prototype to production deployment.
02
Featured Work
Multi-agent voice coaching platform
A real-time voice AI platform that simulates interview, sales, and learning scenarios — with internal agent handoffs, streaming speech interaction, and structured evaluator feedback. Deployed on Cloud Run with a WebSocket voice backend.
Real-time speech-to-speech interaction
Multi-agent orchestration with handoffs
Structured evaluation and scoring
Production backend on GCP Cloud Run
03
Approach
From use case to production
Most engagements follow a phased path — short feedback loops, visible progress, and direct tradeoff discussions.
01
Define the use case
Scope system boundaries and architecture options against real product context.
02
Build a working prototype
Validate interaction quality and technical feasibility before larger rollout.
03
Validate and refine
Tighten evaluation loops and remove reliability issues found under real usage.
04
Deploy to production
Cloud infrastructure, integration, monitoring, and a path to ongoing operation.
“Before LLMs, I spent twenty years building real-time media and AI systems across broadcast, streaming, and analytics. That foundation is why these systems are designed for reliability, not just demos.”
About the background →Exploring an AI product direction?
I’m open to selective conversations on prototype builds, architecture reviews, and AI system design for adjacent product problems.