Hi, I'm Winson Wu.
12 years' experience in software development, including 6 years specialising in computer vision products and AI systems. End-to-end engineering experience across enterprise AI projects, covering architecture design, development, and production deployment. Skilled in model engineering, large-scale distributed systems, and edge AI, turning algorithmic capabilities into stable, scalable, production-grade systems.
Chinese AI company (IPO in 2023; approx. NZD 5B market cap)
Led development of multiple enterprise CV platforms and products, covering requirements analysis, system architecture, and core module design, and driving products from 0 to 1 through continuous iteration.
Responsible for the large-scale production rollout of enterprise CV products, integrating AI models with business systems and addressing challenges in large-scale deployment, performance optimization, system stability, and on-site adaptation.
Built an end-to-end model iteration pipeline covering camera integration, data feedback, annotation, model training and evaluation, and automated deployment.
Ranked Top 20% in performance reviews for multiple consecutive years; co-inventor on 9 CV-related patents.
Recruitment website
Optimized the retrieval architecture and query performance, reducing query latency and improving system throughput.
Remote sensing data platform
Developed modules for managing and analysing remote sensing image data.
Responsible for the day-to-day maintenance and technical support of IT infrastructure, servers, networks, and security systems.
May 2026
A privacy-aware photo retrieval and reasoning system combining SigLIP2, hybrid vector and BM25 search, local VLM reranking, and multimodal answering.
April 2026
An end-to-end macro reconstruction workflow spanning controlled image capture, COLMAP camera tracking, data preparation, and 3DGS training on Apple Silicon.
February 2026
A fully offline species classifier that combines CNN predictions with geographic context and runs on a Raspberry Pi Zero 2 W.
July 2025
A neural attenuation field architecture combining multi-resolution hash encoding, residual U-Net blocks and channel attention.
June 2026
A production deal intelligence platform using multimodal LLMs, model routing and caching to structure text and images, with self-healing pipelines that recover from source changes.
Multimodal LLM / Model routing / Self-healing pipelines
July 2026
A browser-native PII filter that detects and redacts private data across text, PDF, Word, and Excel. Everything runs on your device.
On-device AI / WebGPU
August 2026
Detects and removes visible corner watermarks using computer vision and inpainting, with side-by-side comparison for review.
Computer Vision / Inpainting
University of Canterbury
Dongguan University of Technology
