eisonAI
Local-first reading system for macOS / iOS / iPadOS / Safari Extension
- Built a shipped Apple product combining Safari Extension and app experiences, with on-device inference via WebLLM and MLCSwift.
- Designed long-document routing and chunking around real token-budget and context-window constraints.
- Productized the Cognitive Index concept to make structure visible before content and reduce reading friction.
- Implemented bundled model assets and privacy-first local workflows with CloudKit sync.
Syncnext
Long-running tvOS media runtime
- Built and maintained a tvOS media product over multiple years under real-device constraints.
- Designed split runtime boundaries between native playback/focus control and external high-volatility source/plugin layers.
- Solved problems across HLS/transmux, playback recovery, focus-system design, and Apple TV platform limitations.
- Turned a non-typical app on tvOS into a maintainable runtime with plugin boundaries and operational resilience.
HLN Machine
White-box local AI video pipeline
- Built a fully local, restartable multi-stage AI pipeline that turns a news seed into a completed short-form video.
- Structured the system around IR.json, checkpoints, dependency hashes, partial reruns, and human-reviewable intermediate states.
- Worked across LLM, VL, TTS, ASR, video generation, subtitle correction, and automation on macOS.
- Focused on observability, controlled failure recovery, and white-box design instead of black-box magic AI workflows.