Three-way classification
Detects non-hate, hate, and sarcastic Punjabi Gurmukhi text.
Major Project · NLP · Deep Learning
Alfaaz
Punjabi Hate Text Detection System
A clean AI interface for reviewing Punjabi Gurmukhi text and identifying whether a message is non-hate, hate, or sarcastic. Built for fast moderation workflows, transparent confidence, and practical real-time use.
Live Classification PreviewPOST /predict
Input stream
Waiting
Model confidence
--
{ "status": "initializing" }
Product Analyzer
Check Punjabi Text
The result uses the model’s returned probabilities directly. Alfaaz only improves how the confidence is visualized for users.
alfaaz_review.py
Non-hate
Hate
Sarcastic
Ready to review
Enter a Punjabi comment to see verdict, confidence, hateful profile, and response time.
Capabilities
Focused. Clear. Useful.
Alfaaz avoids over-explaining the research and keeps the interface centered on the actual moderation decision.
Transparent confidence
Shows the winning label and the exact probability spread returned by the model.
Real-time API
Connects directly to the deployed inference service with timeout and offline states.
Punjabi-first UI
Uses proper Gurmukhi typography for input, examples, and classification review.
Trust Signals
Compact Model Snapshot
A concise production snapshot with the signals users actually need: evaluation quality, supported labels, script focus, and live inference behavior.
88.4%
Validated Accuracy
Held-out evaluation score for the Punjabi classifier.
0.883
Weighted F1
Balanced metric across non-hate, hate, and sarcastic classes.
3
Safety Labels
Safe, mild concern, and hate/sarcastic review states.
Gurmukhi
Input Script
Optimized for Punjabi text written in Gurmukhi script.
Integration
Use Alfaaz Anywhere
The frontend is product-facing. The API remains accessible for technical use and future dashboard integration.