I'm the engineer customers talk to: I learn what they need, build it, and stay until it runs in production. Mostly that means LLM agents, the evals that make them trustworthy, and the infrastructure underneath.
Now
- Founding Engineer at D2 Labs. Sole engineer on two ed-tech platforms that deploy LLM agents in the public sector. MyVoice is used by ~20,000 educators across six states. Students First took North Dakota's statewide school system from kickoff to go-live in about a month.
- Nearly Useful, my indie game studio. I'm training an experimental 81M-parameter rectified-flow model that animates characters from game state at runtime. Debut title FINAL OUT! ships this fall.
Before
- Co-founder & CTO, New Memories. Built a multi-agent platform on GCP, plus a system that monitors ~100k physical locations across New York using small tuned models on self-hosted GPUs.
- Capital One, Associate → Lead. Worked on the control plane of the enterprise ML platform (scheduling, distributed training, serving) for a few thousand data scientists.
Things I reach for: Python · Go · TypeScript · Rust · Swift · Kubernetes · Terraform · AWS · GCP · PyTorch · MCP
Things I think about: agent evals and guardrails · context engineering · inference cost and latency · making AI systems a customer can actually trust