7. AI Vision Pipeline
- Each background job shall send the image to a vision API (e.g., OpenAI Vision or Google Vision) requesting labels and confidence scores.
- The system shall discard any label below 0.40 confidence.
- The remaining labels shall form the AI Vision Descriptor list for that image.
- The server shall combine Enhanced Keyword Set + AI Vision Descriptor into a ranked pool, giving 70 % weight to Vision tags and 30 % to user keywords when resolving conflicts.
- The final pool shall be truncated to a maximum of six distinct, natural‑language tokens.
No due date
0% Completed
No results
Try adjusting your search filters.