NLP Pipeline

UC-AIML-002: Intent Classification Pipeline

Purpose: Accurately categorize user messages into actionable intents.

Property	Value
Actor	System
Trigger	Message arrival
Priority	P0

Main Success Scenario:

1. Preprocessing (tokenization, stop-word removal).
2. Vectorization (embedding generation).
3. Inference against trained classifier (e.g., Random Forest/BERT).
4. Confidence score calculation.
5. If confidence > threshold, route to handler; else, fallback.

Alternate Flows:

• Flow A: Low Confidence: Trigger human handoff or clarification-seeking response.

Acceptance Criteria:

1. [ ] F1 Score > 0.85 on test set.
2. [ ] Inference time < 200ms.