I'm finishing a Master of Applied Computing at Lincoln University and looking for a summer internship in New Zealand from November 2026. I build computer vision and document-search systems. Before that I spent 8+ years commissioning robots and wafer-handling automation in Singapore, Taiwan and India, so I know how these systems behave on a real factory floor.
Currently advancing through the Master of Applied Computing at Lincoln University, New Zealand, my core focus is applying software engineering, computer vision and machine learning to real systems for New Zealand industries.
Through Lincoln COMP639 Agile Studio and independent research, I design and ship applications: Kiwi Sorter (YOLOv8 + Detectron2 AR inspection), Rōpū Mapi (standalone Android APK with native voice navigation), SmartShop (Canterbury grocery PWA with Gemini Vision) and DocMindAI (hybrid RAG search engine).
That work sits on 8+ years of industrial engineering across Singapore, Taiwan and India: automation systems, AMHS cleanroom handling, Universal Robots and MiR AMRs at STMicroelectronics (ST STAR Award 2024 for a 30% throughput increase), Tata Electronics and Solomon Technologies.
Vision models fail differently on a factory floor than in a notebook: lighting shifts, parts arrive misaligned, and a stopped line is expensive. I've spent eight years on that floor. Now I build the models too.
Grounded in shipped repositories, unit tests and factory-floor work.
Model fine-tuning, defect segmentation & RAG architectures
APIs, mobile APKs, databases & a GitHub Actions APK build
AMRs, collaborative robotics, AMHS & precision motion
Architecture, tests and source code for each system.
YOLOv8s detects fruit from a webcam, OpenCV GrabCut measures width, height and area in centimetres (37 px/cm), and the Claude Vision API grades each fruit across 10 fields. Bounding boxes are anchored on real fruit through XReal One AR glasses. Two Mask R-CNN models were fine-tuned separately in Detectron2.
Open-source convoy navigation app built as a standalone Android APK with a GitHub Actions workflow. Features a four-layer native TTS voice bridge, speed limit alerts, auto-rerouting, and solo navigation without a backend server (routing and map tiles still need a data connection).
Canterbury-tested grocery PWA with OpenStreetMap Overpass retail store discovery, Traveling Salesperson multi-store shopping route optimization, and Gemini 3.5 Flash receipt vision, backed by 85 passing unit tests.
Document intelligence engine combining FAISS dense vector search with BM25 keyword matching via Reciprocal Rank Fusion (RRF). Features scoped document querying, inline citations, SQL analytics, and Docker deployment.
Self-designed mobile manipulator combining MiR autonomous base with Universal Robots arm. Features a ~30-minute rapid swap, a URCap interface and built-in safety interlocks. Sold and delivered 3–4 units, used for machine-to-machine transfer and aircraft-engine visual inspection at Pratt & Whitney Singapore.
Teleoperated hexarotor crawler engineered to land on live high-voltage power transmission lines and traverse conductors using rubber-coated rollers. Pan-tilt camera streams wireless video for ground inspection without helicopters or climbing.
8+ years delivering measurable engineering and fab efficiency gains.
Master of Applied Computing at Lincoln, NCKU Taiwan, and Veltech India.