Skip to main content
Version: Next

Post-Application Tracking

The Tracking Inbox monitors Gmail for job-application responses and updates timelines.

Overview​

  1. Scans Gmail for recruitment-related emails
  2. Matches emails to tracked jobs using AI
  3. Updates timeline/state when confidence is high
  4. Queues uncertain matches for manual review

Smart router flow​

flowchart TD
    A[Recruitment email arrives in Gmail] --> B[Smart Router AI analyzes content]
    B --> C{How confident is the match?}

    C -->|95-100%| D[Auto-linked to job]
    D --> E[Timeline updated automatically]

    C -->|50-94%| F[Goes to Inbox for review with suggested match]

    C -->|<50%| G{Is it relevant?}
    G -->|Yes| H[Goes to Inbox as orphan]
    G -->|No| I[Ignored]

Setup​

Prerequisites​

  1. Gmail account with application emails
  2. Google OAuth credentials

Configure OAuth​

Set:

GMAIL_OAUTH_CLIENT_ID=your-client-id.apps.googleusercontent.com
GMAIL_OAUTH_CLIENT_SECRET=your-client-secret
GMAIL_OAUTH_REDIRECT_URI=https://your-domain.com/oauth/gmail/callback

Then connect in UI via Tracking Inbox → Connect Gmail.

Using the inbox​

  • Review pending items in Tracking Inbox
  • Approve to link/update timeline
  • Ignore to mark non-relevant

Confidence interpretation:

  • 95-100%: auto-processed
  • 50-94%: pending review with suggestion
  • <50%: pending review as orphan/ignored

Privacy and security​

  • Scope requested: gmail.readonly
  • Minimal metadata sent for matching
  • Email data stays local in your instance

API reference​

MethodEndpointDescription
GET/api/post-application/inboxList pending messages
POST/api/post-application/inbox/:id/approveApprove message
POST/api/post-application/inbox/:id/denyIgnore message
GET/api/post-application/runsList sync runs
POST/api/post-application/gmail/connectStart OAuth flow
GET/api/post-application/gmail/callbackOAuth callback

Common issues​

  • No refresh token: disconnect and reconnect Gmail.
  • Emails not appearing: check runs, OAuth config, and recruitment subjects.
  • Wrong matches: expected in lower-confidence buckets; use manual review.