Video summary

AI Agent Recruiter with n8n: Hire Smarter in 40 minutes!

Main summary

Key takeaways

Technology

Overview

The video demonstrates building an AI Agent “recruiter/head-hunter” pipeline in n8n that:

  • Takes a job brief
  • Scrapes candidates from LinkedIn (via Apify)
  • Uses OpenAI to score and generate a “SWAT-style” analysis per candidate
  • Exports the ranked results to Excel

What the system does (end-to-end)

1) Collect requirements via a form input

The user submits:

  • Candidate keywords
  • Number of candidates
  • Location/Country (e.g., “20 marketing experts in Germany”)

The data is pinned for workflow use.

2) Scrape candidate profiles using Apify (API Actor HTTP request)

  • The workflow calls an Apify actor: LinkedIn Companies/Profiles Bulk Scraper
  • Since this actor isn’t native to n8n, it uses an HTTP request to run it.
  • Execution uses:
    • Run actor synchronously (waits for completion)
    • Then returns dataset items (scraped profiles)

Notes/limitations:

  • The actor can scrape some candidates matching criteria but may not capture all candidates even if you request many.

3) Use an AI step to parse the user query into structured search parameters

Because the user input is one phrase (e.g., “20 marketing experts in Germany”), an AI agent converts it into JSON fields:

  • query (keyword)
  • limit (amount)
  • location (country)

4) Provide the job description from Google Docs

  • The job description is stored in Google Docs.
  • Optimization: fetch the document once (not per candidate) by placing the Google Docs retrieval outside the per-candidate loop.
  • Possible extension: use Google Drive to select different job descriptions depending on role.

5) AI scoring per candidate (SWAT / SVRO analysis)

  • The video references an OpenAI model: GPT-4.1 mini

  • The prompt instructs the model to behave like a professional headhunter, producing:

    • Score (0–100) with a facts-based, “no bias” approach
    • Strengths
    • Weaknesses
    • Risks
    • Opportunities
    • Summary

  • Candidate context passed to the model includes scraped fields such as:

    • name
    • headline
    • about section
    • education (multiple entries)
    • experience (multiple entries)
    • languages (optional)

6) Structured output parsing (JSON)

To make results exportable to Excel, the workflow enforces a structured output parser, so the model outputs consistent JSON keys:

  • name, score, strength, weaknesses, risks, opportunities, summary (Sometimes “summary” issues appear for some candidates.)

7) Export ranked results to Excel

  • Uses the Microsoft Excel node to append rows to an existing sheet.
  • Excel columns map to the JSON output:
    • Name, Score, Summary, Strengths, Weaknesses, Risks, Opportunities

Implementation detail:

  • The model may return arrays; the video notes Excel mapping may not accept arrays directly.
  • The workflow maps specific array indices (e.g., strengths[0..]) into Excel columns.

Performance / scaling claims

  • Demonstrated running for 50 candidates in about 6–7 minutes
  • Rough scaling estimates:
    • ~14 minutes for ~100 candidates
    • ~1.5 hours for ~600 candidates (approximate extrapolation)
  • Final Excel output is sorted by score (highest to lowest).

Practical cautions / observed issues

  • Scraping coverage is imperfect: Apify actor may not scrape all candidates at larger scale.
  • Some candidates may miss summary: even when other fields are present.
  • For direct outreach: you’d need LinkedIn URL fields; the video suggests adding that as an additional scraped field and mapping it into Excel.

Tutorials / guide-like elements highlighted

The video emphasizes how to:

  • Split the user query into structured variables using an AI agent (keyword, limit, location)
  • Call Apify actors via synchronous HTTP + dataset items
  • Avoid re-reading the same Google Doc by fetching it once
  • Enforce structured JSON output parsing for automation compatibility
  • Map array outputs into Excel columns using fixed indices
  • Test with small limits first (e.g., 3–5), then scale up

Main speakers / sources (as stated)

  • Speaker: Max (from Fullite AI)
  • Tools/services referenced:
    • n8n
    • OpenAI (GPT-4.1 mini)
    • LinkedIn
    • Apify (API Actor: LinkedIn profiles/company bulk scraper)
    • Google Docs / Google Drive
    • Microsoft Excel

Original video