Stop routing docstrings to 70B models with on-device AI on Snapdragon | BRKSP90
On Device AI Edge AI Inference routing Model quantization Snapdragon NPU LLM optimization AI cost reduction Microsoft build 2026 Edge inference Model routing
This session from Microsoft Build 2026 demonstrates how to implement a three-tier AI inference routing architecture using Snapdragon X2 Elite's 80 TOPS NPU, routing simple tasks like docstrings to on-device models (≤13B params) while reserving cloud 70B+ models for complex reasoning. Developers building AI coding assistants will learn to cut cloud tokens by 67% and latency by 70% while keeping most code processing local.
Related videos
Stop routing docstrings to 70B models with on-device AI on Snapdragon | BRKSP90
Scale agentic AI from on-device to cloud orchestration | BRKSP92
Taking your AI to the edge with .NET MAUI | OD803
Build automated agents using optimized AI Foundry models on Snapdragon | DEMSP380
Expand local AI reach with Windows ML | OD851
Qualcomm quick take: Live from Microsoft Build | LIVESP126
AI solutions built to power industrial innovation and sovereign control | OD839
Build and deploy AI at the edge for real-world impact | OD837
Run AI SREs without burning token budgets | ODSP928
Context engineering explained: What every AI developer should know
A Context Hack
