job_trackerdemo · fixture data · no inbox read
← landing
applications
10

in your pipeline

this week
0

filed in the last 7 days

in motion
7

applied + interviewing

offers
1

offered + accepted

auto-file gate

0.85

macro-F1

0.979

CI floor

≥ 0.95

01 · pipeline

pipeline distribution · 10 applications40% advanced past applied
applied
4
40%
interviewing
3
30%
offered
1
10%
rejected
2
20%
applied4

  • Cedar Labs

    Software Engineer, Platform

    Your application was received

    filed Jul 8

  • Summit Platform

    Full-Stack Engineer

    Application under review

    filed Jul 10

  • Quarry Data

    Data Engineer

    Thanks for applying

    filed Jul 11

  • Copperline

    Backend Engineer, Payments

    We received your application

    filed Jul 14

interviewing3

  • Northstar Systems

    ML Engineer

    Interview availability — technical round

    filed Jun 28

  • Harbor Analytics

    Backend Engineer

    Next step: online assessment

    filed Jul 2

  • Juniper Cloud

    Infrastructure Engineer

    Schedule your onsite loop

    filed Jun 20

offered1

  • Beacon Health

    ML Engineer, Risk

    Congratulations — offer details inside

    filed May 30

rejected2

  • Fernworks

    Systems Engineer

    Update on your application

    filed Jun 5

  • Atlas Freight

    Software Engineer II

    Moving forward with other candidates

    filed Jun 12

recent activity

latest 5

  • Copperline · Backend Engineer, Payments

    appliedJul 14, 2026

  • Quarry Data · Data Engineer

    appliedJul 11, 2026

  • Summit Platform · Full-Stack Engineer

    appliedJul 10, 2026

  • Cedar Labs · Software Engineer, Platform

    appliedJul 8, 2026

  • Harbor Analytics · Backend Engineer

    interviewingJul 2, 2026

02 · decision trace — click an email to open its verdict

CI-gated at 0.95 macro-F1 · 0.979 measured
1 rules2 e53 SetFit below 0.85 → human

  • Fired at layer 3 · SetFit few-shot head — the learned call

    0%gate 0.85100%

    Confidence clears the 0.85 gate, so Applied files this as “interview” automatically.

Each verdict is traced through three layers — regex rules strike first, e5 embeddings arbitrate, a SetFit head renders the call. Below the 0.85 gate nothing is auto-filed; the email waits for a human and the correction becomes new training data. 0.979 macro-F1 on the committed eval set, gated in CI at 0.95.

Run a full sample inbox through the real classifier