Sample inbox
Eleven realistic job-search emails, each run through the actual Applied classifier. Click any message to trace its path through the three layers and see where it lands against the 0.85 auto-file gate.
How these verdicts are produced. The emails are synthetic — invented companies, no real mail is read. The verdicts are not: each is the exact output of the shipped 3-layer classifier (the same quantized int8 ONNX pipeline running on the Hugging Face Space ↗). Layers 2–3 (the 23 MB model) are precomputed offline; layer 1 (regex rules) is recomputed live in your browser below each trace. Nothing here is hand-tuned.
- scanned
- 11
- auto-classified
- 9 · 82%
- held for review
- 2
- layers
- rules · e5 · SetFit
1 rules2 e53 SetFit below 0.85 → human
- 1 · rulesanswered — applied @ 95% — regex answered
- 2 · e5never ran
- 3 · SetFitnever ran
0%gate 0.85100%Confidence clears the 0.85 gate, so Applied files this as “applied” automatically.
layer 1 recomputed live in your browser:applied @ 95%✓ this is the final verdict
run layer 1 live · your own text
Layer 1 (the regex rules) runs entirely in your browser — no model download, no network. Type anything and watch it score. Layers 2–3 (the 23 MB ONNX model) are the precomputed part above.
verdictinterview95%clears the layer-1 accept bar (≥0.90) — rules would answer
interview +15
Want this on your own mail? Sign in and connect Gmail read-only — the same classifier labels your real inbox.