Three-phase pipeline: a planner breaks down the task, a coder executes edits with tool calls, and a reviewer validates the output. Runs local LLMs on consumer GPU hardware (AMD RX 7900 XT with ROCm). Built because I wanted a coding assistant I could actually trust with my codebase.

Full-stack web app that uses Vertex AI to analyze patent claims and flag potential NPE (patent troll) activity. Built with Next.js, Postgres, and a curated watchlist of 130+ known entities. The kind of project where I learned that legal data is messier than telecom data, which is saying something.

12 years in telecom and cloud infrastructure at Nokia and Verizon, mostly making sure large-scale platforms stay up and building automation so people don't repeat work. These days I'm deep in AI/ML infrastructure: local LLM inference, agentic workflows, and figuring out how to make AI tools that actually work in production.

I like hard problems, ugly debugging sessions, and shipping things that other people use.

B.S. Computer Engineering, Texas A&M ('13). Started at Nokia doing telecom platform integration and customer-facing upgrades. Moved to Verizon's cloud platform team — 8 years on OpenStack, OpenShift, and the automation that holds it all together. AWS Certified Cloud Practitioner. I write Python, Bash, and whatever else the problem needs.

First commit. A Go server and a "coming soon" page held orbit for years.

Rebuilt as a static Astro frame. Hull: S3 behind CloudFront with ACM TLS and a Route 53 docking address. Launch rail: GitHub Actions — Playwright tests gate every deploy, smoke checks confirm orbit. Budget watchdog Lambda kills the CDN if costs ever spike.

Three.js Earth with real day/night shading, cockpit consoles, mobile carousel.

ENG BAY and OBSERVATORY interiors opened. Live telemetry tied in: real mission elapsed time, real ISS ground track, terminator driven by the actual sun.