After every client meeting, someone manually transcribes notes, creates tasks in Monday.com, and emails action items. This takes 2+ hours per meeting and tasks fall through the cracks — 80% of action items had no follow-up within 48 hours.
Three-layer pipeline: n8n orchestrates the workflow (ingest → AI classify → deduplicate → generate tasks → approval gate → route to external APIs). Flask handles the human-in-the-loop approval dashboard. Django stores tasks in PostgreSQL and syncs to Monday.com and GitHub.
Designed and built the entire system end-to-end: n8n workflow, Flask HITL dashboard, Django backend with registration, admin UI, and integrations. Deployed to Railway with PostgreSQL. All three components are production-ready.
The main challenge was deduplication — transcripts arrive in batches with overlapping data. Built a two-pass filter that checks Drive filenames first, then uses Levenshtein matching on transcript content to catch near-duplicates. Reduced false positives by 95%.
Every generated task requires admin approval before syncing to external tools. The Flask dashboard shows pending tasks, lets admins approve/reject with reasons, and auto-approves after 15 minutes to prevent pipeline stalls.
80% faster follow-up on meeting action items. 10+ boards automated. 3 integrated components (n8n, Flask, Django). Tasks now sync to Monday.com and GitHub Issues automatically after approval, eliminating manual data entry entirely.
Loading workflow data...