I built ForgeOps, a Go control plane that uses GitHub webhooks to provision isolated, Helm-deployed preview environments for pull requests on a Kind cluster. Its Kubernetes-based state model pairs a drift reconciler with a worker queue, and I added webhook validation and GitHub Actions checks.
For LLMProxy, I developed an OpenAI-compatible L7 reverse proxy with configurable load balancing, health-check failover, and per-tenant rate limiting. Redis caching brought repeated-prompt responses from 800–2500ms upstream round-trips to sub-millisecond on cache hits, with performance tracked in Prometheus and Grafana.
I also built a serverless document Q&A service using Azure Functions and Azure AI Search, with an event-driven ingestion pipeline for uploaded files. As an AI Model Evaluation Contractor, I evaluated multi-step model tasks and reported recurring workflow failures and defects.
