Hey, I'm Alex

I built this because most similarity recommenders only let users pick a game, then hand back results without explaining what actually matched. If I like Persona 5 for the jazz fusion OST and modern Tokyo setting, I want to search from those reasons, not just a broad RPG label. The goal is to help people find new games, give niche games a better highlight, and show which features drove each recommendation.

A wine recommender developed with the Superlinked team during early access to their SIE engine. It uses document processing, vector embeddings, and small model inference to explain why a result appears, whether the match came from fizz, cherry notes, body, acidity, or other wine attributes.

We built Maldemic to help close the gap between researchers and the public. Disease models can feel locked behind papers and equations, so we turned SIR dynamics and Markov chain mobility into a 3D globe people can watch, question, and reason about. Python computes the stochastic population transitions, then Godot makes the spread visible for public education.

A hackathon project built to make retirement questions feel less foggy. Users can ask things like "Can I retire in the Philippines?" or "How much should I start saving?" and the system answers with retrieved context and visible data instead of pretending a prompt is a financial plan.

A search pipeline built around data movement more than page scraping. 300 extract workers download and parse pages while 300 database workers batch writes into SQLite, with queues keeping the system fed without falling over when the crawler moves faster than the database wants it to.

An open source ML security project that treats training data as the place where model risk often starts. The system uses a multi agent analysis pipeline to inspect dataset content, generate hypotheses, and surface examples worth investigating before bad data becomes expensive behavior.

A gamified Canvas LMS tracker that pulls assignments into a game loop, then scores urgency from completion patterns and difficulty signals. The useful part is turning school workflow data into a next action system students can act on without another dashboard yelling at them.

A hackathon built Godot experiment in systems that can keep expanding. Swirled Perlin noise places planets, wave function collapse handles city placement, and procedural rules create enemies and asteroids, making the project feel like a small galaxy generated from reusable data rules.

A fast game jam pitch: make a tiny character readable, charming, and playable in a single day. We built and demoed Cake Walk at GDC Festival of Gaming with Keriya Son on 3D, Angie Peczon on art, Eric Taylor on shaders, and Ilce Perez on music.

A small vertical climber built over a weekend for a UC Merced game jam in 2022. It is here less as a technical flex and more as the first shipped artifact: a reminder that finishing a small loop teaches more than endlessly planning a bigger one.

A small community plugin for reducing friction while writing in Obsidian. It catches repeated misspellings, applies quick corrections, and keeps a personal dictionary for words Obsidian should stop fighting you on: a tiny version of the same pattern I like, cleaning a messy text stream into something easier to use.

A small dietary search project with a practical pitch: pull in messy menu and nutrition information, normalize it enough to filter, and give people a cleaner way to decide what they can eat.

A compact teaching visualization for spiral growth using queued directions. The pitch is simple: when a system changes step by step, showing the state often teaches faster than another paragraph of explanation.