Multimodal Hub team, building shipped AI systems for Whole-of-Government agencies.
- Rat detection MLOps loop. Cold-started RF-DETR on a manually labelled seed of trap footage, then wired it into a flywheel: new clips in, model auto-labels, human corrects uncertain frames, and the corrected set goes back into training. Less manual labelling was needed each round.
- Smoking detection on public CCTV. Built a cascade detection pipeline that narrows from body to head before classifying smoke vs no-smoke, with ByteTrack and ReID layered on top to identify individual smokers and avoid double-counting across cameras.
- Wildlife detection and species ID. Built the full stack with D-FINE for real-time edge detection, SAM3 and BioCLIP for open-vocabulary species identification, and an iNaturalist reference set localised to Singapore species. Gemini reasons over the output and triggers warnings in real time.
- Vessel mooring compliance for MPA. Built a pipeline that ingests drone imagery of ships at their moorings, detects and identifies each vessel, then checks an approved schedule to flag vessels at the wrong buoy or outside their allowed window.
- Generative documents for public officers. Built a guided workflow that turns PDFs or prose into editable slides, infographics, and diagrams using reference examples, code generation, and image generation. Shipped as a Whole-of-Government proof of concept on PlatformAI.
- NParks claims auto-approval. Extracted structured fields from receipts, invoices, and claims, then matched them through a configurable rules engine so reviewers only handled edge cases.
Product work: built the frontend, backend, and serverless deployment for each POC.
Cloud: AWS Lambda, S3 Vectors, Bedrock Nova embeddings, GCP Cloud Run, Firestore, RunPod, Docker, and GPU-backed inference on DGX Spark.
Additional work: resolved dependency and GPU compatibility issues in existing services and contributed to QA across team deliverables.
