Patrick Semler

Texas A&M CS ’28 · College Station, TX

Software Engineer

I'm a Computer Science student at Texas A&M who likes building software for problems I run into. I work on full-stack products at Finch and research tools at A&M, and I spend my own time building LeetMind and LiftLedger. I still enjoy the same part of programming that hooked me early on: figuring out why something broke and making it work better the next time.

Projects

  • LeetMind

    I built LeetMind because I wanted coding practice that focused on the topics I was actually weak at. It tracks Elo ratings across 20 algorithm concepts and generates problems aimed at each learner's gaps, then checks every generated problem with authored tests, an independent solution, and 50 reproducible randomized inputs before publishing it. Python and C++ submissions run in isolated Docker containers without access to the network, hidden tests, or expected answers.

    • TypeScript
    • Python
    • C++
    • React
    • FastAPI
    • PostgreSQL
    • Docker

  • LiftLedger

    I built LiftLedger to make workout and nutrition tracking feel less like data entry. One dashboard combines Hevy workouts, Apple Health, and meals logged over Telegram, with a muscle-heatmap body chart, goal tracking, and a tool-calling AI assistant. The Telegram logger turns meal photos and text into calorie and macro entries, grounding the values in USDA data.

    • TypeScript
    • React
    • Hono
    • PostgreSQL
    • Vercel AI SDK
    • Vercel

Experience

  1. Research Software Engineer

    2026-08present

    Texas A&M University

    Built a five-stage U.S. beef supply-chain simulator with weighted representative agents that runs 30 million head over 10 years and 250 trials in under 0.6 seconds, fed by an automated pipeline that turns 179,206 rows from four USDA ERS sources into 11 annual profiles.

    • React
    • TypeScript

  2. Machine Learning Research Engineer

    2026-062026-08

    Texas A&M University

    Cut a veterinary classifier's peak memory use by 74%, from 2.6 GB to 689 MB, so it could deploy on Vercel while keeping 90.83% accuracy. Also built a RAG pipeline with Voyage embeddings, Supabase pgvector, and dual-query reciprocal-rank fusion that retrieves the correct source nearly 1.5× as often as single-query search.

    • Python
    • Django
    • scikit-learn
    • Supabase
    • pgvector
    • Gemini
    • RAG

  3. Software Engineer Intern

    2026-01present

    Finch

    Rebuilt onboarding as a resume-first flow alongside a new job board, cutting signup-to-dashboard time ~99% in automated testing, and built a job-matching pipeline that ranks jobs by fit, recency, and network signals with tailored resumes and cover letters.

    • React
    • TypeScript
    • Flask
    • PostgreSQL
    • Supabase
    • Cloudflare Workers

Leadership

  1. Project Lead

    2026-09present

    Product@TAMU

    Lead a team of 5 designing a cross-device feature for multi-device Apple users in an Apple vs. Samsung product case competition between 8 teams.