I build systems software and AI infrastructure — currently shipping production tooling at Electroimpact for aerospace manufacturing automation. My recent work spans metrology systems, distributed inspection tooling, and a pre-deployment safety testing platform for AI agents.
A pre-deployment safety testing platform for AI agents operating in side-effecting business workflows. The platform runs agents under controlled fault conditions — injected timeouts, partial failures, network errors — and uses deterministic invariants to evaluate actual system state rather than trusting the model's self-report. A safe agent supplies idempotency keys so retries are no-ops; an unsafe agent issues duplicate refunds and gets blocked.
Deployed to AWS ECS Fargate behind an Application Load Balancer, backed by RDS PostgreSQL, with infrastructure provisioned via Terraform and a full GitHub Actions CI/CD pipeline using OIDC-based temporary credentials — no long-lived secrets stored in CI.