Education-gap ML paper
Used KNN, SVM, and gradient-boosting models to analyse education gaps between developed and developing nations. Published in IJMR. My first real research paper.
Mostly systems and measurement: serving models, benchmarking them honestly, and building agents that don't fool themselves. Each one started with a question that bugged me. Hover a row for a live preview, or click in for the full story and demos you can play with.
An autonomous ML research agent that knows when its validation set is lying, and refuses to believe it.
Oct 2026 · Agents 02What happens when you stack three "free" speedups on a reasoning model, priced in actual dollars per correct answer.
Jul → Aug 2026 · Benchmarks 03A from-scratch model server with batching, queues, caching, and backpressure. It gets 2.7× the throughput of the naive version and fails politely instead of crashing.
May → Jul 2026 · Systems 04The Huawei paper: per-expert precision for MoE models. 2.9 bits per weight with 4-bit quality. The full story lives on the Work page.
Oct 2026 · Paper 05Won 2nd at Apple Spatial AI Hacks, then kept going. We're now taking it into hospitals.
2026 · Ongoing

Most hackathon projects die on Sunday night. This one didn't. We placed 2nd at Apple's Spatial AI Hacks, and instead of moving on, the team kept building.
Now we're taking it into hospitals to talk with the clinicians and staff who'd actually use it, and to find out what it really needs to be.
Used KNN, SVM, and gradient-boosting models to analyse education gaps between developed and developing nations. Published in IJMR. My first real research paper.
Co-founded a student social-impact venture and built an Arduino device that tracks particulate levels for allergy sufferers, plus the website. Every dollar of revenue went to charity.
As Head of App Dev at GIIS Tech Club, I ran app-dev workshops and organised the 10th GIIS Hackathon for 600+ students across 4 countries in APAC.