Founder, architect and sole developer since 2023 - web, iOS, Android and infrastructure. A marine-life platform for the Red Sea with its own photo-identification service, published mobile apps, and self-hosted production. Most of the underwater photography is my own.

• Built FishID, photo identification against 1,337 species from 10,646 Red Sea photographs, as its own recognition service - dataset shipped as a versioned artefact that updates production without a deploy or migration, decoy classes so a photo that is not a Red Sea fish is refused rather than misidentified, and a single entitlement function resolving admin grants, credit packs, subscriptions, Stripe and Apple/Google in-app purchases
• Extended FishID to video on iOS: scene detection with sharpest-frame selection by variance of the Laplacian, and HDR tone-mapping, so a fish is identified from a shaky dive clip
• Migrated the live public site from client-side rendering to React Server Components with isolated client islands - dive-site page HTML fell from 632 KB to 188 KB, worst cases 2.6 MB to 274 KB
• Raised mobile Lighthouse across every major template (home 65 to 93, dive sites 71 to 90, dive centres 78 to 92), with LCP on the worst pages down from 7.2s to 3.8s and blog CLS from 0.247 to 0
• Moved HTML and image delivery to a Cloudflare CDN with automatic cache invalidation on content change, cutting TTFB from ~1.3s to ~20ms; built a responsive WebP pipeline that cut LCP image weight threefold and removed 136 competing preload requests on a single page
• Shipped native iOS and Android clients with an offline-capable catalogue - both cleared App Store and Google Play review and run in production - including an App Clip, home and lock-screen widgets, an iMessage extension and a share extension that sends a photo to FishID straight from the gallery
• Made the iOS app universal across iPhone, iPad and Apple silicon Mac, with a layout that switches between tab bar and sidebar by measured width, plus Handoff and universal links tying it to the website
• Automated store releases against the App Store Connect and Google Play APIs - versions, localised release notes, builds and submission - each step dry-run by default and refusing to run on a version mismatch
• Replaced SQL pattern matching, which degraded as the catalogue grew, with full-text search for fast fuzzy species lookup, then migrated it off the end-of-life Sphinx 2.2.11 onto Manticore 29 - rewriting the client onto SphinxQL first so the client swap and the engine swap stayed independently revertible, and proving the result with a differential harness that asked both engines the same 27 queries and diffed the returned ids
• Zero-downtime releases via a Docker Swarm stack with start-first, health-gated rolling updates deployed by immutable image digest, after a tag-based deploy silently rolled out nothing and left production on stale code
• Multi-layer caching - Redis, Next.js ISR, and Cloudflare edge purges behind one invalidation path
• Health probe over MySQL, Redis and the search engine, used both as the Swarm health gate and by an automated post-deploy smoke test
• Built a TypeScript analytics ingestion pipeline against the App Store Connect Reports API - ES256 JWT signing, paginated report discovery, gzipped TSV parsing, idempotent daily sync with a 3-day reprocess window

Real-time aircraft tracker that decodes Mode-S and ADS-B signals into a live map and receiver analytics. An MIT licensed Next.js successor to the original Mode-S Mixer.

• Custom Google Maps overlay rendering hundreds of live aircraft without UI lag
• Socket.IO streaming between the dump1090 decoder and the client
• Polar coverage charts and message-type analytics with Highcharts
• Multi-architecture Docker builds (AMD64/ARMv7) for Raspberry Pi deployment