I'm Huston Williams, a computer science student at the University of Houston. I build entire apps from front end to back end. My personal favorite projects that I reccommend you take a look at are the trading platform and the research section which shows data analysis over years of data.
An automated day-trading system that screens the market, runs a portfolio of intraday momentum strategies, and executes trades through the Alpaca brokerage API. An async FastAPI backend streams live market data over WebSockets, persists to PostgreSQL, and auto-starts at the opening bell — paired with a dedicated backtesting suite and a typed React dashboard with live charts.
A Django platform where users sign up, post property listings with photo galleries, and bookmark their favorites. Each listing has a detail page with beds/baths/sq-ft and an interactive map (Amazon Location Service) that pins the property, with images uploaded to and served from AWS S3.
A blogging platform built on Django and backed by DataStax Astra (Apache Cassandra) — exploring a distributed NoSQL database instead of a traditional relational store.
A multi-page Mexican restaurant website with a browsable menu and a working shopping cart — built from scratch with vanilla HTML, CSS and JavaScript.
An interactive tic-tac-toe game built with React 19, using function components and hooks to manage game state, track moves and detect a winner.
I built a research pipeline to design, test and validate intraday trading strategies against roughly a decade of market data (2016–2026). Rather than brute-forcing every minute of every stock, it funnels: the full US equity universe is ranked at daily resolution, narrowed by an hourly screen, and only the day's top momentum candidates are analyzed minute-by-minute for entry conditions. Across 255 strategy variants and parameter sweeps of up to 2,160 configurations.
The funnel is the point: examining every minute of ~7,400 stocks would be millions of bars a day of mostly-wasted compute. Coarse-to-fine screening makes a decade of data tractable and focuses the expensive minute-level analysis where an edge could actually exist.
Six tunable parameters on one strategy produce 1,728 backtest runs. Each run yields a single number — average P&L per month — and this is the distribution of all of them. Of the 1,626 configurations that actually triggered trades, 1,350 were profitable and 276 lost money; a further 102 were so restrictive they never fired a trade and are excluded here. The median configuration returns $131/month against a best of $1,257 — so the top result is the tail of a distribution, not a typical outcome. Mapping the whole space is how you avoid fooling yourself with one lucky config.
7,572 simulated trades from January 2016 to May 2026, aggregated monthly. Fixed $10,000 per trade with no compounding.
Deployed portfolio, 2025–2026. Sorted by total P&L.
Win rate holding across separate periods is the out-of-sample signal.
Exit distribution across all simulated trades.
A fast-paced, high-volume service role held alongside full-time coursework and the engineering projects above — where accuracy, speed and teamwork under sustained pressure are the whole job.
Customer-facing retail across the sales floor, register and stock — my first job while starting my degree.
Led an after-school STEM club, planned and supervised lessons, and guided students through hands-on projects while keeping parents informed of progress.
Shadowed designers and engineers, used 2D and 3D CAD tools to build models, and ran calculations to analyze and optimize processes while following safety standards.
University of Houston — Houston, TX
San Jacinto College — Pasadena, TX
Got an opportunity, a project, or just want to talk shop? My inbox is always open.