skills.rest
NVIDIA
NVIDIAOFFICIAL
Explore all skills by this author →
WebsiteGitHub

medtech-model-evidence-export

Exports sanitized medical AI inference evidence and quality metrics to MLflow.

3.2k|370|Updated Feb 25, 2026
Search Tags:#reproducibility#inference#mlflow#Databricks#NIfTI#medical AI#evidence export
One-click install
npx skills add https://github.com/NVIDIA/skills --skill medtech-model-evidence-export
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: medtech-model-evidence-export
Source: https://github.com/NVIDIA/skills/tree/main/skills/medtech-model-evidence-export
Command: npx skills add https://github.com/NVIDIA/skills --skill medtech-model-evidence-export

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires mlflow, numpy, nibabel, and includes scripts (resource) components.

What problem does it solve?

After running Medical AI inference (such as NV-Generate CT runs), engineers need a safe, reproducible way to record parameters, quality metrics, and provenance in MLflow without leaking patient identifiers or uploading unapproved medical images.

Core Features & Use Cases

  • Sanitized Evidence Export: Mirrors an existing inference result or evidence pack into MLflow with automatic redaction of secrets, patient identifiers, and raw paths.
  • Quality Metrics & Reproducibility: Logs HU statistics, intensity SNR, mask foreground percentage, RNG seed, config digests, and recipe hashes for engineering verification.
  • Controlled Artifact Policies: Supports metadata-only, preview (PNG slices), and full raw NIfTI upload policies, with explicit confirmation required for live medical artifact uploads.
  • Use Case: After an NV-Generate CT inference run completes, preview the export in dry-run mode, review the sanitized parameters and metrics, then log the evidence to a Databricks MLflow workspace with approved slice previews.

Quick Start

Ask your agent to preview what this inference evidence pack would send to MLflow by running the export script in dry-run mode against your result JSON or evidence pack directory.

Frequently Asked Questions about medtech-model-evidence-export

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I export medical AI inference results to MLflow?▼

Run the export_evidence_pack.py script with your result JSON or evidence pack directory, choosing dry-run, local, or databricks mode. Start with dry-run to review sanitized params, metrics, and the artifact plan before any live logging.

How do I preview an MLflow export without contacting a tracking server?▼

Use --mode dry-run with --artifact-policy metadata. The script collects and prints the full export_result JSON contract locally without importing MLflow or sending any data to a tracking server.

Can I upload NIfTI images and masks to Databricks MLflow?▼

Yes, but only with --artifact-policy preview or all in a live mode, plus the --confirm-medical-artifact-upload flag after you approve the destination. Raw image and mask uploads require the all policy and respect --max-artifact-mb.

Does this skill support live training metric tracking?▼

No. This is a post-hoc inference evidence exporter, not a live training-curve tracker. Training workflows should integrate MLflow directly inside their training loops instead.

Why does the exporter fail with 'evidence source not recognized'?▼

The source must be a direct result JSON file, an evidence pack directory containing manifest.json, or a trusted-run root containing skill_run/manifest.json. Pass one of these paths as the positional source argument.

What dependencies are required for NIfTI quality metrics?▼

NIfTI quality metrics and slice previews require numpy>=1.24,<3 and nibabel>=4,<6, while live local or databricks modes require mlflow>=2.10,<4. Dry-run mode works without MLflow installed.

Related Skills

Explore More Data & Analytics Skills→
GoogleCloudPlatformGoogleCloudPlatform
163

gke-inference-quickstart

Automate AI inference deployment on GKE with GIQ and MCP tooling.

Official
Advanced
ivanshamaevivanshamaev
14

mlflow-data-pipelines

Configure MLflow tracking and Model Registry for ETL and training runs.

Community
Advanced
synaptiaisynaptiai
4

send

Enforce policy gates and explicit approvals before transmitting messages to external systems.

Official
Advanced
hung-phanhung-phan
1

experiment-tracking

Track machine learning experiments and manage model lifecycles with MLflow, Weights & Biases, and Trackio.

Community
Advanced