AI Resume-to-Job Matching System (n8n Automation)

Workflow - Step by Step

Webhook Trigger → PDF Text Extraction → Resume Parsing (LLM) → Fetch Matching Jobs → Attach Candidate Profile → Loop Over Jobs → Score Each Match (LLM) → Re-attach Job/Profile Data → Filter by Score → Log to Google Sheets

• Webhook – Receive Resume : A POST endpoint (`/resume-matcher`) accepts an uploaded resume file and triggers the entire workflow.
• Extract Text From PDF : Extracts raw text content from the uploaded resume PDF so it can be sent to the LLM as plain text.
• Groq – Parse Resume : Sends the resume text to Groq's llama-3.3-70b-versatile model with a strict system prompt instructing it to return only JSON matching a fixed schema: name, target_role, years_experience, skills, and domains. Temperature is set low (0.2) and the response is forced into JSON mode for consistent, parseable output.
• Parse Candidate Profile (Code node) : Parses the JSON string returned inside the LLM's response into a clean, usable object for the rest of the workflow.
• Fetch Jobs – Remotive API : Queries the public Remotive remote-jobs API using the candidate's target role to pull currently open, relevant listings.
• Prep Job List + Attach Profile (Code node) : Strips HTML tags from job descriptions, truncates them to a manageable length, takes a batch of listings, and pairs each job with the candidate's parsed profile so both travel together through the rest of the workflow.
• Loop Over Jobs (Split In Batches) : Iterates over the job list one item at a time so each job can be scored individually rather than in one large, unwieldy prompt.
• Groq – Score Match : For each job, sends the candidate profile and job description to Groq with a scoring prompt that returns a JSON object containing a 0–100 score and a reason of no more than 25 words. Temperature is set slightly higher (0.3) for more natural reasoning, still forced into JSON mode.
• Attach Score to Job (Code node) : n8n's HTTP node replaces an item's data with the raw API response, which would normally lose the original candidate/job context. This step re-attaches that original data alongside the parsed score and reason - a workaround for a common n8n data-loss gotcha.
• Filter – Score Threshold : Keeps only matches that clear the configured score cutoff, so only genuinely relevant matches move forward.
• Google Sheets – Log Match : Appends each qualifying match - candidate name, job title, company, score, reason, and application link - as a new row in a shared Google Sheet, then loops back to process the next job in the batch.

Challenge

• Recruiters manually reading through resumes and cross-checking them against dozens of live remote-job boards is slow, inconsistent, and doesn't scale. There was no low-code way to combine LLM reasoning with live job-market data in a single automated, auditable pipeline that a non-engineer could still operate and modify.
• Complexity & Innovation
• Dual-temperature LLM design - a low-temperature (0.2) call forces deterministic structured resume parsing, while a separate, slightly higher-temperature (0.3) call produces more natural, human-readable match justifications.
• Per-job scoring loop instead of batch prompting - each job is scored independently against the candidate profile, avoiding context dilution and keeping every score individually explainable.
• Data-loss workaround - a dedicated 'rehydration' code node solves a common n8n gotcha where HTTP nodes silently overwrite upstream item data, re-attaching candidate and job context after every API call.
• Fully low-code orchestration - the entire multi-stage pipeline (parsing, live job retrieval, scoring, filtering, logging) runs inside n8n with no dedicated backend service to deploy or maintain.

Process

• Defined the webhook contract - a single POST endpoint accepting a resume file upload to trigger the workflow.
• Built PDF text extraction to normalize any uploaded resume into plain text for LLM consumption.
• Designed the resume-parsing prompt with a strict JSON schema and low temperature for reliability.
• Integrated the Remotive public API to pull live, relevant remote job listings by target role.
• Built the batch-loop scoring stage, sending one job + profile pair at a time to Groq for a 0-100 fit score and reason.
• Added the rehydration code step to fix n8n's item-context loss after HTTP calls.
• Applied a configurable score-threshold filter before logging.
• Connected Google Sheets as the final, recruiter-facing output layer.

Complexity & Innovation

• Dual-temperature LLM design - a low-temperature (0.2) call forces deterministic structured resume parsing, while a separate, slightly higher-temperature (0.3) call produces more natural, human-readable match justifications.
• Per-job scoring loop instead of batch prompting - each job is scored independently against the candidate profile, avoiding context dilution and keeping every score individually explainable.
• Data-loss workaround - a dedicated 'rehydration' code node solves a common n8n gotcha where HTTP nodes silently overwrite upstream item data, re-attaching candidate and job context after every API call.
• Fully low-code orchestration - the entire multi-stage pipeline (parsing, live job retrieval, scoring, filtering, logging) runs inside n8n with no dedicated backend service to deploy or maintain.