About
Engineering the layer between models and reliable products
I build the engineering layer that turns models into reliable products. My focus is inference, evaluation, agents, and the observability and cost controls around them.
Software engineer with 8+ years building the systems that turn models into reliable products: model-serving and inference workflows, agentic systems, evaluation loops, data pipelines, observability, authorization, and cost controls. My background is in software that has to work in production, across large-scale distributed systems, cloud-native infrastructure on AWS and Kubernetes, applied optimization, and reliability engineering. I care as much about latency, cost, safety, and evaluation as I do about model quality, and I hold 5 U.S. patents in automated network optimization.
How I work
Principles
Reliability is a feature
Latency, failure modes, and degraded behavior are product requirements, not afterthoughts. That instinct came from resolving 100+ Tier-3 production incidents.
Measure, don't guess
Evaluations and benchmarks gate changes. I'd rather ship a smaller claim backed by a reproducible number than a big one backed by a vibe.
Safe by default
Human approval for risky actions, server-side authorization, audit logs, and prompt-injection tests. Safety is designed in, not bolted on.
Cost is a metric
Quotas, caching, mock modes, and ephemeral compute keep spend bounded and observable rather than a surprise on the invoice.
Reproducible over clever
Config-driven pipelines and idempotent automation over one-off scripts, so results can be re-run and trusted.
Portable at the app layer
Kubernetes and Helm keep the application cloud-neutral; provider-specific choices stay isolated in infrastructure.
Focus
Where I go deep
- LLM inference & serving
- Evaluation & data flywheels
- Agentic systems & RAG
- AI SRE & observability
- Feature platforms & real-time ML
- Cost & reliability engineering