README.md

metadata

title: Prior_Auth_pipeline
app_file: app.py
sdk: gradio
sdk_version: 5.49.1

Prior Authorization — Lightweight MLOps Pipeline (Demo)

This repository is a demo MLOps pipeline for Prior Authorization (PA) use cases.
It is designed to deploy on Hugging Face Spaces (Gradio) with Python 3.10.

Features

• Synthetic data generation (or upload your CSV)
• Preprocessing + training (RandomForest pipeline)
• Model registry (joblib files + registry.json)
• Active model (copied to models/active_model.joblib)
• Inference endpoint (Gradio file upload)
• Monitoring with Evidently (Data Drift report)
• Simple heuristic-based retrain trigger (auto retrains when drift detected)

CSV schema for inference & monitoring

Your CSVs should contain these columns:

• age
• prior_auth_count
• chronic_conditions_count
• severity_score
• cost_estimate

How to deploy

1. Create a new Space on Hugging Face: https://huggingface.co/spaces/LianHP/Prior_Auth_pipeline
   • SDK: Gradio
   • Runtime: leave default (we use runtime.txt to request Python 3.10)
2. Upload files:
   • app.py
   • requirements.txt
   • runtime.txt
   • README.md
3. Wait for the build to finish. Open the Space and try:
   • Train a model
   • Run monitoring
   • Upload your CSV to run inference

Notes

• This is a demo pipeline with synthetic data and simple heuristics.
• For production: integrate a proper model registry (MLflow), CI/CD, secure storage (S3), authentication, logging, alerting, and privacy-preserving synthetic data practices (HIPAA considerations).