Screen CVs with OpenAI and PostgreSQL using chained prompts

How it works

• Webhook receives a job ID and list of candidate IDs from your database
• If the job has no template yet, Prompt 0 reads the job description and automatically extracts mandatory requirements, differentials, behavioral competencies and sets the weight of each criterion
• For each candidate, 3 prompts run sequentially with accumulated context:
  • Prompt 1 scores the candidate (0–100) against the job template using calibration anchors to avoid score inflation, plus a breakdown score per criterion
  • Prompt 2 receives the score as context and identifies strengths with concrete resume evidence, separating critical gaps (missing mandatory requirements) from secondary gaps (missing differentials)
  • Prompt 3 receives the gaps as context and generates personalized interview questions for that specific candidate — not generic HR templates
• Results are saved directly to PostgreSQL after each candidate
• When all candidates are processed, Prompt 4 automatically generates an executive summary of the entire pool with recommendations on who to interview

Set up steps

• Add your OpenAI credentials to all AI nodes (~2 min)
• Add your PostgreSQL credentials to all Postgres nodes (~2 min)
• Create the required tables using the SQL schema provided in the workflow sticky note (~5 min)
• Trigger via POST /webhook/cv-analyze with { "job_id": 1, "candidate_ids": [1, 2, 3] }