posted an update

Final submission: EdgeRoute Arm is live

EdgeRoute now routes AI-agent requests locally on Arm64 before an action runs—without sending prompt text to a cloud classifier.

Measured results:

  • 74.72% smaller INT8 ONNX model
  • 33.1% higher batch-128 throughput
  • 100% FP32/INT8 prediction agreement
  • 20/20 on a separate hand-authored phrasing challenge

I also report the tradeoff: INT8 single-request latency is slightly slower, while footprint and batched throughput improve substantially.

Watch the 60-second demo · View the source and reproducible results

Log in or sign up for Devpost to join the conversation.