Why not put AI inside the scoring system?

Small quantized models hallucinate and are unpredictable. Big models are heavy and need strong hardware (RX 6700 XT 12GB cannot sustain 5 CV parallel). Workflow orchestration (pdf.js + n8n + rule 40/30/30, 70% threshold) is stable, fast (~5s), auditable, and 100% local: AI was only a supervised syntax assistant (Ornith 35B).

How does HR know why a CV passed or failed?

Each row has a clear reason: e.g., Lolos 80% utk Drafter & 3D Visualization or Tidak Lolos 35% utk Junior Architect. The CSV header is Indonesian: Peringkat, Skor (0-100), Status, Alasan, so HR opens it in Excel and knows immediately. Threshold 70% is adjustable.

What if the CV is a scanned PDF or the webhook is offline?

Scanned PDFs with no text layer trigger a toast (text <10 chars) but don’t fail the batch. The item is sent with empty text and an _error flag. If the webhook is offline (15s timeout), the landing page shows a fallback Download JSON so the demo still works.

From 30 Minutes to 5 Seconds: Automating Mass CV Screening Without In-System AI.

Home Profile Experiments Contact Experiment 001 Experiment 002 Experiment 003 Experiment 004 Experiment 005 Experiment 006 Experiment 007 Experiment 008 Experiment 009 Experiment 010 Experiment 011 Experiment 012 Experiment 013 Experiment 014 Experiment 015 Experiment 016 Experiment 017