feat: aggressive augmentation + moderate ridge + final retrain on all data

fix: split before augment to prevent data leakage

feat: data augmentation + smaller model to fix overfitting

fix: inference auto-stops on silence, one-shot predict_proba_sequence

feat: utterance-level inference with predict_proba_sequence

feat: strengthen ridge regularization (alpha 0.01~10)

revert: remove moving-average smoothing

feat: moving-average smoothing (last 5 probs)

fix: match inference normalization to training (normalize → trim → re-normalize)

fix: singular matrix fallback to lstsq + raise min ridge_alpha

Upload requirements.txt with huggingface_hub

Upload app.py with huggingface_hub

initial commit