Refresh onnx/ export to current 39-label model (adds messages.read; matches flat top-level)

Refresh pytorch/ to current 39-label model (adds messages.read; matches flat top-level)

Update train.json to current intent set (39 labels, 1699 examples; adds messages.read + expanded system.oos/tests/symptoms)

Retrain: recall/meta + general questions -> system.oos (2026-06-25); refresh model card

retrain: +24 vital-normal phrasings -> tests.test_history (offline L9 parity)

Push model using huggingface_hub.

v2: fix 8 mined misroutes (meds->routines, doctor->appointments, meal-smalltalk->oos, not-feeling-well->symptoms, email->profile read); +46 examples; held-out acc 0.99

Retrain: greeting/help as system.oos (engine-robustness); acc 0.98 / macro-F1 0.97

Retrain: terse messaging + care-team + bare-name oos (1472 ex, holdout 1.00)

Retrain to 38 labels (+meal_time.read, +meal_time.update, +summary.read); fill model card [train data]

Retrain to 38 labels (+meal_time.read, +meal_time.update, +summary.read); fill model card [pytorch mirror]

Retrain to 38 labels (+meal_time.read, +meal_time.update, +summary.read); fill model card

chore: training data — 1259 examples, 38 classes

chore: ONNX export — summary.read retrain

chore: retrain — summary.read + system.oos + meal_time (38 classes)

Add SetFit intent router model (BAAI/bge-small-en-v1.5, 35 intents)

initial commit