Play 88 β Visual Product Search π
AI visual commerce β image-based product matching, multi-modal search (image+text), shoppable image recognition, visual similarity ranking.
Build a visual product search engine. CLIP/Florence encodes product catalog into vector embeddings, AI Search provides sub-200ms vector similarity search, multi-modal fusion combines image + text queries (βblue version of thisβ), shoppable images detect individual products in lifestyle photos, and click-through feedback continuously improves ranking.
Quick Start
cd solution-plays/88-visual-product-search
az deployment group create -g $RG -f infra/main.bicep -p infra/parameters.json
code .
# Use @builder to implement, @reviewer to audit, @tuner to optimizeArchitecture
π Full architecture details
Pre-Tuned Defaults
- Encoding: CLIP ViT-L/14 Β· 768-dim Β· background removal Β· center crop
- Search: HNSW index Β· cosine similarity Β· threshold 0.65 Β· top 10 results
- Reranking: Visual similarity 50% Β· in-stock 15% Β· popularity 15% Β· price 10% Β· recency 10%
- Multi-modal: Image 70% / text 30% Β· text overrides color/size/material
DevKit (AI-Assisted Development)
| Primitive | What It Does |
|---|---|
agent.md | Root orchestrator with builderβreviewerβtuner handoffs |
copilot-instructions.md | Visual search domain (CLIP embeddings, multi-modal, shoppable images) |
| 3 agents | Builder (gpt-4o), Reviewer (gpt-4o-mini), Tuner (gpt-4o-mini) |
| 3 skills | Deploy (210+ lines), Evaluate (120+ lines), Tune (225+ lines) |
| 4 prompts | /deploy, /test, /review, /evaluate with agent routing |
Cost Estimate
| Service | Dev/Test | Production | Enterprise |
|---|---|---|---|
| Azure AI Vision | $0 (Free) | $250 (Standard S1) | $800 (Standard S1) |
| Azure OpenAI | $25 (PAYG) | $350 (PAYG) | $1,200 (PTU Reserved) |
| Azure AI Search | $0 (Free) | $500 (Standard S2) | $1,000 (Standard S3) |
| Container Apps | $10 (Consumption) | $200 (Dedicated) | $500 (Dedicated HA) |
| Cosmos DB | $3 (Serverless) | $150 (2500 RU/s) | $450 (8000 RU/s) |
| Blob Storage | $2 (Hot LRS) | $50 (Hot ZRS) | $200 (Hot GRS) |
| Key Vault | $1 (Standard) | $5 (Standard) | $15 (Premium HSM) |
| Application Insights | $0 (Free) | $40 (Pay-per-GB) | $120 (Pay-per-GB) |
| Total | $41/mo | $1,545/mo | $4,285/mo |
π° Full cost breakdown
vs. Play 09 (AI Search Portal)
| Aspect | Play 09 | Play 88 |
|---|---|---|
| Focus | Text-based hybrid search | Image-based visual search |
| Query | Text keywords | Upload photo + optional text |
| Encoding | text-embedding-3-large | CLIP ViT-L/14 visual encoder |
| Use Case | Enterprise document search | Retail product discovery |
π Full documentation Β· π frootai.dev/solution-plays/88-visual-product-searchΒ Β· π¦ FAI Protocol
FAI Manifest
| Field | Value |
|---|---|
| Play | 88-visual-product-search |
| Version | 1.0.0 |
| Knowledge | R2-RAG-Architecture, F1-GenAI-Foundations, O2-AI-Agents |
| WAF Pillars | performance-efficiency, reliability, responsible-ai, cost-optimization |
Last updated on