Field service engineer who builds the tools.
Four years keeping production critical semiconductor equipment running in a live fab. I build the custom systems that turn analog signal into digital record, with a receipt for every fault. That is the overlap I work in: now building the tools I wanted on the floor, and checking them against the published literature instead of against a good feeling.
Open tools for data center power service: switchgear, protective relays, ATS, metering, commissioning.
The rule: nothing ships until it reproduces a published worked example and the number matches the book.
| Live | What it proves |
|---|---|
| Power chain one line explorer | Five published UPS configurations, utility to rack. Click any component to take it out of service and watch what goes dark. Every assertion quotes the source paper's own words. It refuses to print a tier, on purpose: a tier is awarded against a standard I have not read, and a confident wrong answer is exactly what this repo exists to avoid. |
Nine more in the same discipline: coordination curves, ATS sequence, relay bench, Modbus metering, ladder logic, switching orders and LOTO planning, NETA test plans, commissioning scripts.
That gate is not decoration, and it cost me things. Mutation testing my own test suite (breaking the
code on purpose to prove the tests could actually fail) showed one of my features was dead code, so I
cut it. The same pass caught a real bug: a bus tie wired one way, so one side could never back feed the
other. ./tools/gate.sh re-proves all of it in one command.
Four years of field service on production critical semiconductor equipment in a live fab. Better than 95% uptime, 15 to 25% downtime reduction, average service response under two hours. PVD, CVD, CMP and FSS certified. Trained junior techs.
- Safety. LOTO trained and recertified every six months across those four years, in a live production fab. NFPA 70E training, ongoing, through a U.S. DOL registered electrical apprenticeship.
- Power gear. Hands on across the chain, acquired one install at a time: breaker populated power distribution cabinets feeding an entire tool, plasma process and peripheral sensors both, with the plasma supplies run as master and slave pairs into a single chamber for 40 kW to the plasma. The high power supplies, RF generators, and UPS banks downstream of them. Protective relays I set, tested, replaced, and chased nuisance trips on. Ran the power and signal cable that ties all of it together, with the interference discipline that goes with it: power kept away from signal, and loop area kept small. Verified components against their published norm values, worked out why they failed, and recommended the fix.
- Controls and comms. Migrated tool host communications off legacy SECS-I serial onto HSMS over TCP/IP, across many tools, each one bringing its own legacy quirks. Chased SECS/GEM failures that were corrupting tool behavior analytics down to the board. RS-232 and Modbus TCP across tool and facilities systems: vacuum, thermal, gas, RF, and plant equipment. Worked the fab's facility supervisory layer, the monitoring PC that manages the tool computers, the peripheral sensors and the control valves. The vendor shipped that box closed, so I got it to log sensor voltage and current and hand the readings to a server over Ethernet, which is how the experiment underneath got its data. Install level networking: bringing every new tool online with remote engineering and IT.
- Before the fab, two years in microelectronics prototype development. Debugged and characterized custom electronics at board and component level: probes, scope, function generator, LCR meter, microscope. Built custom data acquisition fixtures to pull current, voltage, RF and accelerometer data off a board, and turned the results into fingerprints that tell a hardware fault apart from a software configuration fault. I brought that lens into the fab, which is why a tool that would not talk got its comm board characterized instead of swapped.
- The method that actually travels. A problem is a story of the device's state, and every deviation from norm has a hard reason backed up by data. On equipment nobody memorizes, you derive it from the documentation, you engage the experts who know it, and you verify before you trust it. That is what I was paid for, and it is what the toolbox above is made of.
A fab runs 24/7 and production critical, on the same discipline a data center floor runs on.
Edge AI Portfolio · Live demos, no sign in
Verifiable AI tools for nonprofits, on the same instinct as the toolbox: the model observes, and deterministic code you can audit makes the decision.
| Project | What it proves |
|---|---|
| GrantMatch | Reads a funder's RFP into a checklist of every requirement, then refuses to mark a draft compliant unless it can quote your own words back. Catches the model fabricating coverage, live. |
| ColdWatch | A roughly $21 edge sensor that predicts a food bank fridge failure hours early, ignores door openings so the alarm stays trusted, and alerts fully offline. Zero AI at runtime. |
| ReClaim Vision | Grades donated computers against a real refurbishment standard. The model observes, a rubric decides. Currently in QA. |
| BenefitsBridge | Answers SNAP and energy assistance questions only from official rule text it retrieved, quoting the rule every time. Ask about a program it has not loaded and it refuses and routes you to a caseworker. English and Spanish. |
Keystone · a living atlas of the world's industrial value chains, built to find the chokepoints. Educational, not investment advice.
Python, JavaScript, MATLAB, C, Java, Verilog, VHDL. Everything above runs in a browser or fully offline, at zero cost per month.
Registered electrical apprentice on the Master Electrician track, 2029. B.S. Business Administration, 2027. A.S. Engineering Transfer. A.S. Mathematics.