ProjectIntent Over Keywords: From Stitched Embeddings to a Production Fashion Agent · Module 11: StyleMe: the product problem · part 13 of 26
Part 13 · Module 11: StyleMe: the product problem
Module 11: StyleMe: the product problem
23 min read·
About 45 minutes · Needs: Module 10 (measuring v1 against v2) · Code: originals/styleme/fashion-agent/app.py, part3-styleme/backend/src/styleme/agent/prompts.py, part3-styleme/backend/src/styleme/agent/planner.py, part3-styleme/backend/evals/datasets/intent_cases.jsonl, part3-styleme/backend/data/
Where this fits
This module (highlighted) opens Part 3: the shopper's chat page, the planner that decides first, the catalogue it searches and the original fashion-agent it grew from (grey). Blue is our system throughout.
Parts 1 and 2 answered one question over one video: find the right moment, then say what happened. Part 3 changes the job. StyleMe is a fashion shopping assistant over a catalogue of 203 products, with text, photos and a conversation that runs over several turns. This module states the product problem before any architecture: what shoppers actually type, why many of their messages must not trigger a search, what the catalogue holds, and what the original fashion-agent got wrong. The high-level design in Module 12 and the modules after it then build and run the answer.
The rest of this project is yours to keep
This project is bought on its own, once, and stays readable afterwards, including the parts added to it later.