Patrick Semler

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

patricksemler@tamu.edu

I'm a Computer Science student at Texas A&M, currently doing machine learning research on veterinary diagnostics. I build full-stack AI systems at ApplyFinch, and before that extended backtesting and risk infrastructure at Maroon Fund. On my own time I'm building LeetMind, an adaptive coding-practice platform that aims each problem at what you are actually weak at, and Phobos, a self-hosted AI chief of staff that can safely write and deploy new tools for itself.

Projects

  • LeetMind

    An adaptive coding-practice platform that generates original problems tailored to each user's per-concept mastery across 20 data-structure and algorithm categories, so every session lands at the edge of their ability. Nothing is published until it survives differential testing between the reference solution and an independent oracle across authored cases and 50 seeded randomized inputs, and submitted code runs in throwaway, network-disabled Docker containers that cannot see the hidden answers.

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

  • Phobos

    A self-hosted AI chief of staff for one person — natural-language requests from a Telegram bot and a React dashboard become approved, durable task graphs, worked by 8 specialized agent roles spanning research, planning, execution, verification, repair, and independent review. When a task needs a capability it lacks, a self-extending forge generates, reviews, builds, tests, and installs a new TypeScript module, then runs it inside a networkless, read-only container that can reach host services only through a grant-checked broker — and nothing executes until the plan and its permissions are approved.

    • TypeScript
    • Node.js
    • Fastify
    • React
    • PostgreSQL
    • Docker

Experience

  1. Machine Learning Research Assistant

    2026-06present

    Texas A&M University

    Diagnostic classification across 143 veterinary conditions — cut the classifier's peak memory by 74% while holding accuracy at 90.83%, and grounded Gemini care plans in each patient's own context instead of the disease name alone.

    • Python
    • scikit-learn
    • Gemini
    • RAG

  2. Full-Stack Developer

    2026-01present

    ApplyFinch

    An AI job-application platform built by a multi-developer team — shipped full-stack workflows for scraping, ranking, and document generation, and hardened the async pipeline against 4 failure modes it used to lose work to.

    • React
    • TypeScript
    • Flask
    • PostgreSQL
    • Cloudflare Workers

  3. Quantitative Developer

    2026-022026-05

    Maroon Fund

    Backtesting and risk infrastructure — added Jensen's alpha, OLS beta, and tracking error against a 3-source SPY fallback, and raised backtest fidelity by modeling slippage, commissions, and forced exits.

    • Python
    • pandas
    • Plotly
    • pytest