ProjectIntent Over Keywords: From Stitched Embeddings to a Production Fashion Agent · Module 24: What changes in production · part 26 of 26
Part 26 · Module 24: What changes in production

Module 24: What changes in production

23 min read·
About 40 minutes · Needs: Module 7 (Gemini Embedding 2 and models as settings), Module 20 (evaluating the agent and its latency gate), Module 23 (the Cloud Run deploy) · Code: part3-styleme/backend/src/styleme/llm.py, part3-styleme/backend/src/styleme/config.py, part2-video-rag-unified/src/core/retry.py, part2-video-rag-unified/src/embedders/gemini_embedder.py, part3-styleme/backend/src/styleme/sessions.py, deploy/README.md, .env.example

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
The parts this module is about: the Gemini models (amber), the Cloud Run deploy and the agent evals (blue) that tell you when production behaviour has changed.

This is the last module. Every earlier one built, measured or packaged something. This one is about what happens after the code works: the model you depend on disappears, the free tier says "too many requests", a long video turns into hundreds of billed calls, and a second server instance would quietly break the cap of 3. All of it happened, or was designed against, while this project was built. The model is amber; everything around it is engineering you can test.

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