Context
Full audit of the job pipeline system. The filter was too aggressive, excluding real AI roles at Sierra, Harvey, Anyscale, and Bay Area locations like Redwood City and Foster City. 254 roles were stuck at outreach_drafted with no nudge mechanism. No Discord or Sheets integration.
What Changed
Filter Fixes (highest impact)
- Location: Expanded allowlist with 12 Bay Area cities + united states, usa, anywhere
- AI-native companies: 30 companies (Sierra, Harvey, Anyscale, ElevenLabs, etc.) now auto-pass domain filter without inflating fit scores
- Director PM: Director of Product Management correctly detected as PM role (was being rejected as Not a PM role)
New Features
- Stale outreach nudge: Daily digest now surfaces roles stuck at outreach_drafted for 7+ days. New pipeline stale CLI command.
- Data cleanup: Auto-archives REJECTED/ARCHIVED jobs older than 30 days to recoverable JSONL
- Discord #top-jobs: New channel with webhook, posts new strong-fit roles daily
- Google Sheets sync: 305 active jobs synced to Pipeline Auto tab in Job Search spreadsheet
- CLAUDE.md: Full system documentation for the pipeline
Test Coverage
219 tests passing (was 169). +50 new tests covering filter fixes, stale reports, resume variant selection, cover letter generation, and outreach drafting.
Key Decisions
- AI_NATIVE_COMPANIES separate from Tier 1/2: bypasses domain filter but does not boost scoring
- Removed us from location list (matched Austin). usa and united states cover it
- Created Pipeline Auto sheet tab to avoid overwriting hand-curated Pipeline tab
- Discord self-service: created channel + webhook via bot API instead of asking user
Open Items
- Re-filter 130 excluded roles with new filter logic
- Triage 254 stale outreach drafts
- Configure google-api-python-client for automated daily Sheets sync
Commit: cd231e4 on assortedLLMTasks/main, 12 files, 781 insertions