Email Classifier for Job Applications
A fine-tuned BGE-small model that classifies emails into job application categories. Designed to run entirely in the browser using Transformers.js.
Model Description
- Base Model: BAAI/bge-small-en-v1.5 (33M parameters)
- Task: 5-class email classification
- Format: ONNX (opset 14, IR version 7)
- Size: 32.5 MB (quantized) / 127.6 MB (full)
Labels
| Label | Description |
|---|---|
confirmation | Application received/confirmed |
rejection | Application rejected |
interview | Interview invitation |
offer | Job offer |
not_job | Not job-related |
Training
- Method: Curriculum learning (2-2-1 epochs)
- Phase 1: 2 epochs on full-body emails
- Phase 2: 2 epochs with 4:1 full-body:snippet mix
- Phase 3: 1 epoch with 1:1 balanced mix
- Training Data: ~28K emails (original + augmented snippets)
- Validation Accuracy: 100% (full-body), 100% (snippet)
Usage with Transformers.js
import { pipeline } from '@xenova/transformers';
const classifier = await pipeline(
'text-classification',
'mattohan/job-tracker-email-classifier',
{ quantized: true }
);
const result = await classifier('Thank you for applying to the Software Engineer position...');
// [{ label: 'confirmation', score: 0.99 }]
Files
├── config.json
├── tokenizer.json
├── tokenizer_config.json
├── vocab.txt
├── special_tokens_map.json
└── onnx/
├── model.onnx # Full model (127.6 MB)
└── model_quantized.onnx # Quantized model (32.5 MB)
Privacy
This model runs 100% client-side in the browser. No email data is ever sent to a server.
License
MIT