Cata is an interactive natural language processing playground and text classification tool. As you type sentences or technical requirements, the classifier highlights key words in real time and predicts relevant category tags alongside exact confidence percentages.
Key Features
- Live Token Highlighting: Visualizes which words and phrases in the input text contribute most to the predicted tags.
- Confidence Scores: Displays mathematical confidence probabilities for every predicted category tag in real time.
- Multi-Label Classification: Simultaneously detects multiple overlapping topics (such as identifying both Frontend and Backend aspects in full-stack queries).
- Edge Deployment: Runs on Cloudflare Workers for near-instant inference and zero cold starts.
- Interactive Playground: Clean sandbox interface designed for testing edge cases, prompt variations, and classification sensitivity.
Why I Built This
I wanted a visual playground to test and understand machine learning classification boundaries in real time. Seeing which tokens activate specific predictions makes it much easier to fine-tune datasets, diagnose misclassifications, and build intuition for how NLP models evaluate developer input.
Tech Stack
- Platform: Cloudflare Workers, Edge Network
- Frontend: React, TypeScript, Tailwind CSS
- ML & NLP: Feature extraction, tokenization, multi-label probability scoring