Work

A full sales-procurement system for export clients—quote, order, purchase, ship—folded out of email, sheets, and chat.

An AI skill for product selection—cross-signal validation instead of browsing charts, with explainable structured output.

Ops coordination lived in spreadsheets and chat → a configurable admin, without rebuilding shop-floor control.

Operations, maintenance, and planning on the floor—complex multi-role paths in traditional energy. Map how work actually runs, then ship configurable admin.

Sales could configure what engineering could not build → one set of rules for sellable and buildable config and per-vehicle BOM.

R&D, manufacturing, and quality engineers—sales, engineering, and production spoke different configuration languages.

A managed deep-learning platform: training jobs, model serving, GPU scheduling. For algorithm teams—not to own business-model quality.

Algorithm engineers—the boundary is hosted environment, not business-model quality.

Built an R&D change platform from scratch: parts, materials, configuration, and process changes, with configurable nodes instead of offline coordination.

Used vision in content safety and commerce listing, as a loop AI, humans, and ops could actually run.

Speaker app and voice UI: device management, onboarding, content, and smart home.

Benefits ran as one-off campaigns → distribution ops can reconfigure, and partners can join by a standard.

A Tmall campaign selection admin: structure the assortment pool by category, then add analytics and models.

Daily and big-promotion mechanics: ops config, merchant signup, and consumer UX on one path.