Open-Weight Model Governance

AI Model Registry & Benchmarks

A version-controlled catalogue of specialized healthcare models optimized for low latency, on-device edge inference, and Indian medical NLP tasks.

Model Deployment & Edge Inference Architecture

All models are deployed adhering to Section 18 of the Blueprint. ASR dialect models and triage classification networks are quantized to 4-bit and INT8 formats to allow offline operation on battery-powered tablet kiosks in sub-centres without continuous internet connectivity.

Local Edge Execution TFLite & ONNX runtimes eliminate server round-trips for vital sign alerts and acoustic cough screening.

Zero Cloud Egress Protected patient voice recordings remain buffered locally and are destroyed post-transcription.

Audit Trail Logging Every model version change triggers automated regression evaluation across standard validation test suites.