DistilmBERT v1 (regex v0 fallback)
Intent Classifier.
Multilingual intent classifier for the multi-agent orchestrator. v0 = regex (live, fallback).
Foundation · DistilmBERT v1 (regex v0 fallback) (Apache 2.0)
The honest description.
Multilingual intent classifier for the multi-agent orchestrator. v0 = regex (live, fallback).
How it works.
User message text (multi-language: EN/HI/TA/TE/UR/KN/BN/MR), message tokens, user context state. Primary agent intent (budget_guardian | bill_payment | savings_optimizer | prediction_engine | fraud_sentinel | tax_planner) + confidence 0-1 + is_low_confidence flag when below threshold.
What it learned from, and how we watch it.
Inputs: User message text (multi-language: EN/HI/TA/TE/UR/KN/BN/MR), message tokens, user context state.
Drift baseline not yet documented for this model.
- PR #10: DistilmBERT v1 trainer wired (Apache 2.0).
- v0 regex remains as fallback when v1 endpoint errors.
Every release, traceable.
Data Mukodi holds about you is subject to DPDP §12 erasure (use /consent in the app), and any concern about this model can be raised via [email protected] under DPDP §13 grievance, or via /security.