Overview
OPIC 1.0 is the first production release of the Operations Intent Classifier. This model provides accurate multi-class intent classification specifically tuned for ecommerce customer support scenarios.What’s New in 1.0
- 10 Intent Classes: Comprehensive coverage of common ecommerce support intents
- Sub-60ms Latency: Optimized for real-time inference
- Batch Processing: Support for classifying multiple messages in a single request
- Confidence Thresholds: Configurable filtering of low-confidence predictions
Model Specifications
| Property | Value |
|---|---|
| Model Version | 1.0.0 |
| Architecture | Fine-tuned BERT-based transformer |
| Training Data | Ecommerce customer support conversations |
| Intent Classes | 10 |
| Max Input Length | 2,048 characters |
| Languages | English |
| Average Latency | < 60ms (p95) |
Supported Intents
| Intent | Description |
|---|---|
track_order | Order/shipment tracking inquiries |
cancel_order | Order cancellation requests |
return_item | Product return inquiries |
refund_request | Refund status or requests |
product_inquiry | Product questions |
change_address | Delivery address modifications |
payment_issue | Payment problems |
complaint | Negative feedback |
compliment | Positive feedback |
other | Unclassified intents |
Quick Start
Single Message Classification
Response
Best Practices
- Use confidence thresholds: Set
thresholdto filter out low-confidence predictions - Request top-k results: Use
top_kto get multiple intent predictions for ambiguous messages - Batch when possible: Group multiple messages to reduce API calls

