Live Sports Technology Platform
A best-in-class native iOS matchday app and the low-latency live-sports backend behind it — an opportunity I spotted, then shaped and shipped with a small team at speed.
Led a five-person function and owned the rebuild from a partner-coupled Laravel monolith to a stateless FastAPI + Next.js platform on AWS. Beyond the platform, the consumer matchday app went from our initial product sketch to the App Store with a small team, a lean budget and a rapid turnaround. On matchday the platform ingests tens of thousands of Opta XML snapshots in near-real-time (~1–3s feed-to-product) and serves complex live data and assets to tens of thousands of concurrent fans at a 5.9ms p95 — validated at 30,020 concurrent users with a 0% error rate.
Music Discovery Platform
An operating system for music discovery — type a sentence, get tomorrow’s roster: 8M+ artists, refreshed daily, learning each user’s taste.
Founded and built playlistn — an operating system for music discovery. A&R teams connect their Spotify and streaming accounts once; the platform ingests their world, learns their taste from every play, save and signing, and keeps tracking live as new artists catch their attention. Discovery becomes effortless: search in plain English, get a shortlist in seconds, explore a living map of 8M+ artists refreshed daily — and carry the whole journey, found to signed, in one tool. Underneath: live multi-source ingestion, a self-growing artist graph, and a custom AI search engine tuned so precision stays high at a fraction of a cent per query. Bootstrapped solo on £30k.
ClaimyRoyalty Anomaly Detection Service
Two-tier royalty anomaly detection: fast inline detection at ingestion time, and deeper nightly automated investigations.
Fractional engagement alongside Baller League. Paired with the CTO to develop technology strategy and roadmap, then built a two-tier anomaly-detection system. Layer 1 — "fast detections" — runs deterministic SQL rules embedded directly in the royalty-processing pipeline at statement-ingestion time, catching known error patterns with zero added latency. Layer 2 — "smart detections" — runs nightly on Cloud Run, aggregating the full time-series picture and running a comprehensive ML-based classification framework to surface anomalies that only become visible at scale or over time.
Royalty Intelligence & Forecasting
Forecasting framework, anomaly-detection system, and statistical modelling for ~45,000 songs.
Built two proprietary data products for music publishing and recording revenues — a forecasting framework deployed in three months and an anomaly-detection system that identified $120M of outlier royalty income on first run.
A&R Recommendation Pipeline
Explainable ML, temporal features, scalable analytics, governed data-platform contribution.
Progressed over four years from royalty operations into analytics and data science, leading the discovery and MVP delivery of an explainable A&R recommendation pipeline and contributing to a governed data platform supporting ~100 engineers and analysts. Kobalt’s engineering standard is exceptionally high — I was fortunate to be mentored there by some incredible data and technology leaders, and everything since is built on those foundations.
Supervised learning, advanced algorithms, unsupervised learning, recommenders & RL