A voice agent you call and talk to: streaming speech-to-text, access-filtered hybrid retrieval, an LLM, cloned-voice synthesis and a lip-synced video avatar. About a second from the end of your sentence to hearing the reply, with barge-in.
Voice backend for a Unity game whose companion character reacts to the world you build. Streaming faster-whisper, Llama 3.1 on vLLM and WebSocket TTS, with time-to-first-audio cut from about 7s to about 1.5s.
Centralized management for public and private Kubernetes clusters through a single dashboard: operators that automate authentication and authorization, secure kubeconfig generation, and a multi-tenant monitoring stack built on OpenTelemetry, Thanos, ClickHouse and Perses.
A Kubernetes operator built on Open Cluster Management that centralizes authentication and authorization across many clusters from a single hub.
GitOps-driven automation for enterprise OpenShift clusters: Ansible playbooks for provisioning and configuration, ArgoCD for declarative state reconciliation and continuous deployment.
Go service for a cloud application marketplace: asynchronous installs and uninstalls, billing, and a publisher/consumer workflow backed by PostgreSQL and Redis.
Full-stack food and money donation platform connecting donors with people in need, with request tracking and role-based access.
Several of the systems below are client or employer work, so there's no public repository: those write-ups describe the architecture and the reasoning without shipping anyone's code. Where a repository does exist, it's linked from the case study.