Confidence-Based Routing: When to Trust Your Classifier
How to use confidence scores and prediction margins to build three-tier routing: auto-route, suggest, or escalate
Text classification without GPUs, API keys, or prompt engineering. Benchmarks, integrations, and things we've built with 15ms classifiers
How to use confidence scores and prediction margins to build three-tier routing: auto-route, suggest, or escalate
Side-by-side comparison of classifying text with an LLM vs frozen embeddings + logistic regression. Same task, both approaches, with code and benchmarks
Benchmarking 16 models on text classification: LLMs, decision models, and classifiers. On 77-class problems, embedding + logistic regression wins by 14+ points
Add deterministic text classification to n8n with Jeffy. Free, self-hosted, 15ms per classification. No LLM, no API key
Add deterministic text classification to Activepieces. Free, self-hosted, 15ms per classification. No LLM, no API key
Regex + weighted scoring beats every LLM on the Jevman Pac-Man benchmark. 8,035 mean score, 3ms per decision, $0
A real-time pattern detector learns your left/right choices and predicts what you'll pick next. How unpredictable are you?