I build the software that diagnoses your Tesla before you arrive.
Technicians need to know what's wrong with a car before it shows up at the service center. I build the automated diagnostics that make that happen, covering every Tesla model and running thousands of times a day. I started as a mechanic, taught myself Python, and now I build AI systems that do the investigating for me.
Vehicle sensor data comes in, anomalies get flagged, and diagnostic workflows fire automatically. This is what production-scale automation looks like.
I own 5 subsystem autodiags in production, covering all vehicle models with end-to-end component diagnostics. I've also been building AI tools that automate the boring parts of my job, things like investigating regressions, fixing classification issues, and writing test cases.
I got pulled into the autodiag world after diagnosing a wiring issue that caught the right person's attention. At the time, autodiags were built in a visual drag-and-drop editor that fell apart for anything complex (try doing a for loop or timestamp comparison in a flowchart). When the team championed script-based autodiags, I taught myself Python through YouTube and Udemy and started building from scratch, because most subsystems had zero automated diagnostic coverage.
I traveled to service centers and the factory to help with critical backlogs and expedite diagnostics. Later moved into a remote specialist role doing the same type of work but from a distance, which is when I started seeing how much of this process could be automated.
I drove across Texas doing mobile service at customers' homes and ranches, sometimes 5+ hours from the nearest service center. If you didn't know exactly what parts to bring, the whole trip was wasted. That's where I started caring about remote diagnostics.
I passed all 8 ASE certifications and specialized in electrical diagnostics. Schematics, multimeter troubleshooting, systematic diagnosis. This is where I got the foundation for everything I've built since.
I wrote 5 subsystem autodiags from scratch, covering all Tesla vehicle models. Each one diagnoses an entire subsystem end-to-end, not just individual components. Before these existed, technicians had to manually test every component. I also converted 100+ existing autodiags from a visual drag-and-drop editor to Python scripts, because the visual editor fell apart for anything complex (try doing a for loop or timestamp comparison in a flowchart).
I built a goal-seeking agent framework that orchestrates AI coding sessions, scheduled jobs, and persistent loops until resolution. Each objective follows a loop: measure, investigate, remediate, re-measure. Some tasks are straightforward, like finding flagged tickets, investigating the root cause, and submitting a fix. Others are more involved, like monitoring diagnostic precision and launching a full investigation when it drops below threshold.
I built an AI-powered dashboard for autodiag developers. Most dashboards just give you numbers. This one gives you the reasoning behind the numbers, and actionable buttons to do something about them: launch a deeper investigation, file a ticket, or add a new test case directly from the dashboard.
I designed an LLM orchestration that takes a customer's concern and pulls from multiple data sources: repair history, fleet-wide failure distributions, and technical reference material. The model could also call tools on its own to run additional diagnostics or fetch more context. The POC didn't land internally, but the architecture was solid and I learned a lot about where LLMs fit (and don't) in diagnostic workflows.
2x backyard competition winner. Offset smoker, mesquite wood only. I organized the Meat Assassins Brisket Cook-Off for three years. 14 hours minimum on a brisket or don't bother.
I ran my first marathon in December 2025, finished at 4:29:46. Already signed up for the next one, going for sub-3:30.
I want to buy 100-400 acres in the Hill Country someday. No rush.
I'm always up for talking about engineering, AI, or brisket.