Senior AI Engineer | Computer Vision, Multimodal & Edge AI
junxian@portfolio ----------------- foundation : production vision + edge AI domain : computational photography adjacent : reliable agent + multimodal workflows runtime : Python | C++/Android | TFLite/ONNX | GPU method : perception -> reasoning -> runtime -> evaluation -> handoff location : Singapore
I build AI systems that have to survive runtime constraints, long-tail visual failures, integration, evaluation, and production handoff.
My production depth is in camera AI, computational photography, edge deployment, and native runtime integration. My public agent work makes the adjacent capability inspectable through tested provider-neutral orchestration, validated contracts, fallback, review gates, and traces.
- Reliable Agent Workflow - provider-neutral routing, retry and fallback, validated output contracts, human-review gates, deterministic traces, and executable hard cases.
- Photography Mentor - browser-local image diagnostics, retrieval, practice planning, review state, and evidence-aware coaching.
- Selected Systems - public-safe case studies across vision deployment, computational photography, and reliable workflow design.
- CV - concise, sanitized screening version.
- Perception: segmentation, matting, monocular depth, detection, tracking, and multimodal inputs.
- Reasoning: product rules, workflow state, retrieval, validated outputs, and review policy.
- Runtime: C++/Android, GPU pipelines, edge deployment, services, and provider interfaces.
- Evaluation: hard cases, visual comparison, traces, failure taxonomies, and release gates.
- Handoff: explicit assumptions, interfaces, debug paths, acceptance criteria, and operating knowledge.
Applied AI Engineer, Forward Deployed Engineer, Computer Vision Engineer, Multimodal AI Engineer, and ML Systems Engineer roles where model behavior, runtime constraints, evaluation, and delivery have to work together.