AI agent engineer · Forward deployed · San Francisco Bay Area

I build agent systems that earn their place in production.

I turn complex business workflows into dependable voice and AI agents—building the runtime, integrations, evaluation loops, and customer delivery systems that take them from concept to value.

Explore selected work Read my experience

Forward-deployed engineering at the intersection of agent architecture, customer delivery, and product reality.

Runtimetools, service wrappers, and observability

Reliabilityevaluation, root-cause analysis, and safeguards

Deliverycustomer ownership, speed, quality, and time to value

Selected work

From prototype to production.

I work where agent capability meets the constraints that make a system useful: business logic, backend systems, real customers, quality, and measurable outcomes.

01 — AI agent engineering

Building voice-agent systems from runtime to rollout

Build Lambda-based agent services that expose tools, prompts, and business logic to production voice agents—then take ownership of the customer implementation around them.

Tool adapters, contracts, and backend integrations
Conditional tool invocation and agent behavior design
Deployment quality, delivery speed, and time to value

02 — Reliability engineering

Making agent behavior inspectable and improvable

Use observability, evaluation, and root-cause analysis to understand failures, detect hallucinations, and make production behavior more reliable over time.

03 — Platform & delivery

Closing the gap between customer needs and the platform

Build reusable SDK integrations and customer-specific workflows while partnering across product and engineering to turn ambiguous needs into durable systems.

See projects and experiments

How I work

Useful beats impressive.

Start with the workflow

Good AI begins with a real user problem, a clear decision, and an honest definition of success.

Design for failure

Guardrails, fallback paths, and evaluation are product features—not afterthoughts.

Ship, measure, improve

The first deployment creates the learning loop. The work is making that loop fast and safe.

Let’s build something useful

Have an AI problem worth solving?

Start a conversation