ProjectIntent Over Keywords: From Stitched Embeddings to a Production Fashion Agent · Module 12: High-level design · part 14 of 26
Part 14 · Module 12: High-level design

Module 12: High-level design

22 min read·
About 45 minutes · Needs: Module 11 (StyleMe, the product problem) · Code: part3-styleme/backend/src/styleme/api.py, part3-styleme/backend/src/styleme/agent/graph.py, part3-styleme/frontend/nginx.conf.template, docker-compose.yml

Where this fits

WHERE THIS FITS Intent Over Keywords this module PART 1: STITCHED VIDEO RAG three models, two spaces, RRF PART 2: UNIFIED VIDEO RAG one model, one space PART 3: STYLEME FastAPI, LangGraph, React QUALITY AND OPERATIONS evals, Docker, Cloud Run Frames ffmpeg, one per 5 s CLIP ViT-B/32 frames + text, 512 d Whisper base speech to text MiniLM 64-word chunks, 384 d Two collections p1_visual, p1_text Rank fusion RRF, k = 60 Gemini answer from transcript text API + Streamlit upload, ask, sources Windows ffmpeg, video + audio Embedding 2 Gemini, one vector One collection p2_unified, no fusion Notes + answer Gemini, cited times React web chat, photo, chips nginx one URL, /api proxy FastAPI /api/chat, limits Sessions in memory, cap of 3 Planner intent before retrieval Implementer batched embed, search Generator stylist reply, ids Grounding check no LLM, drops invented Gemini flash-lite, embedding 2 Qdrant styleme_products Catalogue ingest image + title vectors Langfuse optional traces Test video 26 questions, 4 kinds Video eval Part 1 vs Part 2 Agent evals intent, grounding Guards uploads, limits, secrets Containers compose profiles Cloud Run StyleMe only CI/CD GitHub Actions, WIF Original code v1, v2, fashion-agent
This module covers all of Part 3 (highlighted): the browser, the front door, the API with its sessions, the agent's three nodes and grounding check (green), Gemini (amber, a model), Qdrant, the catalogue and optional tracing.

Module 11 described what a shopper needs from StyleMe: a greeting answered as a greeting, a vague wish turned into one good question, a clear request turned into a search. This module draws the whole system that does it: which processes run, how one chat turn travels from the browser to Gemini and Qdrant and back, and where state lives. Later modules zoom into single boxes.

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.

Or open everything with a plan