Self-taught developer and hands-on car guy in Dallas, headed for mechanical engineering.
I build and repair my own cars: a 2008 Mk5 GTI and a 2016 Camry sleeper.
I build and self-host my own software, including a chat client that starts answering in about 40 ms.
I play competitive tennis for my school team and in UTR tournaments.
Two cars I diagnose, repair and modify myself. The GTI is the project. The Camry is the sleeper.
Mk5, 2.0 L turbo FSI, six-speed manual, high mileage.
Drag the slider to compare the two at any rpm.
| Measure | stock | target | gain |
|---|---|---|---|
| power hp | 200 | 320 | +120 |
| torque lb-ft | 188 | 300 | +112 |
Stock is Volkswagen’s published 200 hp and 207 lb-ft. Target is modeled from the 320 hp and 350 lb-ft goal for the turbo build, not a dyno sheet.
2008 GTI with the six-speed manual and high miles.
A tighter, more precise shift for the six-speed.
Grip first. Tires are the cheapest traction upgrade there is, and everything else builds on them.
An aftermarket intake, plus upgraded ignition coils. Coil packs are a common failure point on the 2.0T FSI.
Took weight out of the car. Less mass helps acceleration, braking and turn-in all at once.
Looked hard at AWD and RWD conversions and weighed nitrous. Chose traction and a coherent turbo setup instead: parts that work together, and power the front tires can put down.
An Integrated Engineering Stage 1 tune, then turbo, fueling and intercooling planned as one package, aimed at roughly 320 hp and 350 lb-ft.
The sleeper. The 2.5 L 2AR-FE looks and runs stock, and inside is a 10-speaker custom audio system I set up myself. Power goals, through ECU tuning, stay reliability-first.
Tracing evaporative emissions faults from the code to the part that is actually leaking.
Diagnosing and repairing air conditioning, which matters in a Dallas summer.
Cleaning carbon off direct-injection intake valves.
Reading live data and codes before buying a single part.
I recreated Volkswagen’s EA888 turbo four in Automation, the engine-design sim, and I spend plenty of time in Forza.
Things I’ve built and run myself, on hardware I manage.
A lightweight multi-lane chat client: a local Qwen model on my RTX 4090 and DeepSeek V4.1 in the cloud. It routes each message to the right lane, keeps long-term memory in SQLite, and has a SwiftUI iOS app.
Measured September 21, 2026 on the RTX 4090: 40.1 to 43.6 ms from request to first streamed token, four warm runs, full server path. Routing decisions take under 10 ms.
A terminal coding agent. It runs on your own machine, reads, edits and searches files, and runs commands, asking before every edit or command. Tachyon is only its model gateway: an OpenAI-compatible API that holds the keys, picks the lane and enforces each user’s daily budget.
files and commands stay here
only model calls happen here
A Jarvis-style personal assistant that lives in Telegram, built on Hermes. It remembers context between conversations, runs automations, and has a routing-debug mode that shows how each request was handled.
I run my own hosting under truexsystems.com: a VPS running Node, and my RTX 4090 machine serving Tachyon and its local model through Cloudflare Tunnel, behind Cloudflare Access. Smaller builds live there too, like a Christmas gift-list site I coded.
Competitive play on my school team and in UTR-rated tournaments.
Nov 2025 to Sep 2026, 10 months
Runner, Turn Close
As a runner, I got food and drinks to guests in the theater during showings. On turn-close shifts I reset auditoriums between shows and closed out at the end of the night.
current
Aiming for mechanical engineering
UT Austin is the target. Most of what I know about software and cars so far is self-taught, from shop manuals, forums, datasheets and a lot of trial and error.
Questions about a build, a project, or an opening? Reach me here.