I build AI-native products from interface to infrastructure, combining mobile architecture, backend systems, and practical product judgment.
Built a GAAP-like multi-currency financial system as a stress test for LLM-assisted development. The backend used strict type boundaries, interface contracts, transaction rules, and property-based testing to keep accounting behavior predictable across double-entry flows.
Delivered AI platform customization for regulated enterprise environments. The work included a high-availability SSO gateway, role-based access control, directory synchronization, and an intranet deployment model that protected private data with minimal upstream changes.
Engineered a production RAG pipeline for a consumer AI app in psychological dream analysis. Dynamic model routing, localized embeddings, custom Chinese tokenization, and semantic caching helped balance response quality, latency, and operating cost under constrained compute.
Built a mobile-first drag-and-drop poster editor for theatrical arts use cases. A hybrid rendering model paired fluid client-side canvas interaction with backend image composition, eliminating device-specific export inconsistencies.
Early work included migrating legacy hybrid mobile apps to native architectures, reverse-engineering undocumented systems, improving unstable communication modules, and establishing CI/CD pipelines for testing, internal distribution, and app release workflows.