Zhijing · AI content workbench
Local workbench
Connect product facts, style settings, editing and human review with export paths.
- My work
- Led workbench development and validation
- Validation
- Partial checks; generation and publishing have not passed end to end
Context
Fabric businesses need content based on product images, specifications and account style. Human review remains part of the flow to keep content aligned with product facts.
My role
I led requirements and workflow design, used AI-assisted development, and handled testing and maintenance.
Work delivered
The workbench covers setup, style confirmation, product input, editing, human review and export. The image shows product input.
Approach
Product facts, content capabilities, human approval and output paths are organized and validated separately. A successful account connection is not treated as evidence of a complete generation flow.
Delivery & feedback
August 2026 records cover a text probe and account login, restart, disconnection and login again. The full-content test failed at delivery of its first image; end-to-end generation and RED publishing have not passed.
Technical approach
- TypeScript
- Node.js
- Provider abstraction
- Schema validation
Product facts and style input → content generation and editing → human approval → export or upload paths. Provider abstractions organize different capabilities.
Implementation flow
- Product facts
Confirm facts and style
- Provider calls
Coordinate generation and editing
- Human approval
Check content against product facts
- Output paths
Verify export and upload separately
Validation scope
Local test and account connection records exist. Successful connection does not establish an end-to-end generation or publishing result.
Talk about your business
You do not need a complete solution. Share one time-consuming task and we can discuss where AI or software could help.