Production web applications, client infrastructure, and applied machine learning.
Consumer food safety mobile app scanning 70k+ UK grocery products for instant allergen verdicts. Engineered for supermarket aisles with offline-first indexed datasets, sub-second barcode resolution, and 14 EU allergen profile matching. Pitch finalist at Bethnal Green Ventures.
Automated multi-format PDF & DOCX text extraction pipeline using pdfjs-dist and mammoth to enrich AI personalisation. Features a 23-step dynamic onboarding state machine with localStorage persistence and GPT-4o prompt guardrails.
End-to-end meeting intelligence pipeline for ChatSDK. Ingests Whereby recordings, executes multi-provider LLM summarisation with background polling & caching, and persists structured action items via Drizzle ORM and PostgreSQL.
High-throughput machine learning pipeline predicting molecular properties and compound toxicity with TreeSHAP explainability and partitioned Parquet data streaming.
Client-side WebGL utility that packs discrete PBR material channels (Roughness, Metallic, AO, Height) into single 32-bit RGBA textures, cutting browser VRAM by up to 75% at 60 FPS.