JobTracker hybrid classifier โ€” SetFit layer

The few-shot ML layer of JobTracker's 3-layer email classifier (rules -> e5-small-v2 similarity -> SetFit). Classifies job-pipeline emails into stages: applied, assessment, follow_up, interview, offer, other, pending_application, rejection.

  • CI-gated at a 0.95 macro-F1 floor; 0.979 measured on the committed v3 evaluation set, alongside a 182-test suite.
  • An int8-ONNX export of this model powers the in-browser demo (zero servers), verified output-identical to this pipeline.

Part of the JobTracker project by Ayush Yadav.

Downloads last month
12
Safetensors
Model size
22.7M params
Tensor type
F32
ยท
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Space using yadava5/jobtracker-setfit-classifier 1