Laolu's Projects
Never manually send a photo again. PhotoLayer is the social layer for the photos app, automatically sending photos that contain friends and creating buzz around the moments you've forgotten.
Popular Apps
Built an AR platform that recreates real dorm spaces, starting with freshman housing at Stanford. Users can walk through these rooms digitally, interact with the space, and personalize them through decoration and layout ideas.
Complete, reproducible pipeline that trains a face-recognition model on a race-balanced corpus, reweights training by continuous skin tone (ITA) instead of coarse race labels, distills it into a smaller model for edge deployment, and evaluates its performance on RFW dataset.
A research tool for measuring how far the tibia rotates during an ultrasound stress test. Used by Stanford's Med school in a novel, ultrasound-based ACL diagnosis tool
A full-stack platform where students study with a multilingual TA voice, while teachers orchestrate learning with file‑ingest, auto‑grading, email blasts, Zoom breakdown, and class management. I owned the end‑to‑end architecture: streaming speech, session orchestration, prompt tools, safe eval, and scalable data flows.
Two resilient scrapers that normalize Nike/StockX catalogs into analytics‑ready CSVs. I built robust selectors, rate‑limit handling, backoff, and schema validation for downstream pipelines.
Terminal game with timers, scoring, and post‑game analytics. I designed efficient word lookups and metrics that surface skill deltas and streak dynamics.
Fully playable terminal Wordle and a solver that achieves >95% success via entropy‑guided guesses and pruned candidate sets. I tuned heuristics for both speed and accuracy.
Campus‑scale lost & found with pgvector image search and CLIP/Gemini verification. I implemented the Supabase schema, auth, vector indexing, and a moderation pipeline that flags fraud in real‑time.
Playable terminal Camel Up with Monte‑Carlo advice. I implemented probability aggregation and expected‑value payouts to guide optimal bets per leg.
Train against algorithmic opponents with leg‑by‑leg feedback. I implemented enumerative hand evaluation and experiment‑based rollouts for situational edge estimates.
Prototype forecasting stack with feature engineering and execution scaffolding. I built modular signal evaluation and a backtest harness; projected ~≈ $150 PnL/min under optimistic assumptions.
Interactive March Madness simulator with varying levels of difficulty