Edge-AI deployment plan: recommended model, expected throughput, fallback stack — avoid 2x hardware overspend.