This repository hosts the trained
residual model bundle for the
nutriSafe system.
The bundle is
not a standalone predictor: it learns a bounded
correction in
[-3, +3] on top of a deterministic clinical rule baseline.
See the
Space for
an interactive demo, or the upstream repository for the full inference
pipeline.
1from huggingface_hub import snapshot_download
2local = snapshot_download("SandipSingh/nutrisafe-residual-xgb")
3
4import xgboost as xgb, joblib, json
5booster = xgb.Booster(); booster.load_model(f"{local}/model.json")
6calibrator = joblib.load(f"{local}/calibrator.joblib")
7schema = json.load(open(f"{local}/schema.json"))
For inference you also need the rule-baseline engine and feature
pipeline — see the upstream repository.
