Resume fit score against a job post

Resume fit score against a job post

An n8n workflow that rates a resume against one job post. Someone uploads a PDF resume on a form and pastes the job
post's link. The workflow reads the resume, fetches the post and has OpenAI pull the description out of the page. An
OpenAI agent then lists the candidate's strengths and weaknesses, rates the risk and the reward of hiring them, and
gives a fit score from 0 to 10 with its reasons. The candidate's name and email come out of the resume, and everything
lands as one row in Google Sheets.

What is inside

• workflow.json: the workflow as n8n exports it, 14 nodes, one of them a sticky note. It holds no credential and no
  sheet ID.
• README.md: these notes.

What we changed from the original copy

• Append Data pointed at its author's sheet. It now asks for YOUR_SHEET_ID and writes to the first tab.
• The Strengths and Weaknesses columns joined each list with join("nn"), because the original copy lost its
  backslashes. They are back, so the items sit on their own lines again: join("\n\n").
• The agent's model node named o4-mini. OpenAI's deprecations page, read on 2 November 2026, lists it as shut down on
  23 October 2026 with gpt-5.6-terra as the replacement, so requests to it fail. The node now uses gpt-5.6-terra
  and carries that name.

Before you start

• An n8n instance, on n8n Cloud or your own server.
• An OpenAI API key.
• A Google account and a sheet for the results.
• Resumes as PDF files with real text in them, not scans.
• Job posts at addresses that open without a sign-in.

Set it up

1. In n8n, create a workflow, open its menu and choose Import from File. Pick workflow.json.
2. Add an OpenAI credential and select it in gpt-5.6-terra and OpenAI Chat Model.
3. Make a sheet whose first tab has these headers in row 1: Date, First Name, Last Name, Email, Strengths,
   Weaknesses, Risk Factor, Reward Factor, Justification, Overall Fit.
4. Add a Google Sheets credential in Append Data and replace YOUR_SHEET_ID with the ID from the sheet's address,
   the part between /d/ and /edit.
5. Open On form submission. Its title is Check Your Resume Score. Rename it if only your team will use it, and set the
   Resume field's accepted file types to .pdf.
6. Save, open the form from its test address, upload a resume with a job link, then publish the workflow.

How a run goes

1. On form submission takes the resume file and the job post's link. Both are required.
2. Extract from File2 reads the text out of the PDF, and Set Resume keeps it as resume.
3. On a second branch, HTTP Request fetches the job link, and Information Extractor1 has gpt-4.1-mini pull the job
   description out of the page.
4. Merge and Aggregate put the two texts together in one item, the resume first.
5. Recruiter Agent compares them with gpt-5.6-terra. Its answer has to fit the six fields in
   Structured Output Parser: strengths, weaknesses, risk with the worst case, reward with the best case, a
   whole-number score out of 10, and the reasons.
6. Information Extractor reads the resume again for the first name, the last name and the email.
7. Append Data adds one row to the sheet, with the date and time of the run.

Change the look

• What the agent looks for: the system message in Recruiter Agent, where the output format and the scale are set.
• Which fields come back: the schema in Structured Output Parser. Change the Append Data mapping to match.
• More contact details, such as a phone number: add an attribute in Information Extractor and a column for it.
• The models: gpt-5.6-terra for the rating and the contact details, OpenAI Chat Model for the job post.
• The form: its title, its labels and its accepted file types in On form submission.

Good to know

• Billing: OpenAI bills three calls a run. The one that reads the job post receives the whole page's HTML, which can
  be long.
• The date: Append Data writes it with hh:m a, so minutes under ten lose their zero, as in 03:5 PM. Change it to
  hh:mm a.
• Files: only PDFs can be read. A Word file stops the run at Extract from File2, and a scanned PDF gives the agent an
  empty resume to rate.
• Job links: many job boards block plain requests or need a browser to show the post. Then the run stops at
  HTTP Request, or the description comes back empty. A company's own careers page works better.
• Contact details: all three fields are required, so when a resume has no email, the model may still return one.
  Check that column before you write to anyone.
• Repeats: every submission adds a new row, so the same candidate sent twice shows up twice.
• Open form: once it is published, anyone with its address can send files and spend your OpenAI credit. The candidate
  sees no score. It goes only to your sheet.
• Models: OpenAI's page lists gpt-4.1-mini for shutdown on 14 January 2027, with gpt-6-luna as the replacement.
  Change OpenAI Chat Model before then.
• Fair hiring: the score is one model's reading of two texts. In some places, New York City among them, software that
  helps rank candidates needs a bias audit and a notice to candidates. Check the rules where you hire, and have a person
  read every row before a decision.
• Your data: each resume and job post goes to OpenAI, and names and emails sit in your sheet. Tell candidates how their
  resume is handled.