AI scored my job application 95/100 before I even saw it (free n8n workflow)

Thirty-one n8n nodes read job boards and LinkedIn every morning, score each new listing against a real CV, and draft the cover letter for anything worth applying to. A schedule trigger reads a Settings tab in Google Sheets for every number the workflow runs on, downloads the CV from Google Drive once and extracts its text, then pulls fresh listings from RSS job boards and from LinkedIn through an Apify actor, normalizes both into one shape, hashes every listing's URL, and drops anything already logged before a single row gets scored. Survivors go to Gemini one at a time with the job listing and the full CV, scored zero to a hundred with one sentence of reasoning. Anything at or above the threshold gets a specific cover letter referencing real details from both the job and the CV; everything else still gets logged for review. Every scored job lands in an Applications tab, and a match at or above threshold posts to Slack with a preview of the letter. Two ideas worth stealing. Deduplication used to mean one sheet lookup per job inside the loop, and it got slower every time the applications tab grew; moving it to a single bulk read before the loop means one sheet call no matter how many jobs get checked. And the AI's score gets pulled out with a plain regex instead of a structured output parser, because the structured parser's overhead was enough to crash the run outright on this instance. I walk through every node. Want the workflow file? Comment "jobs" and I will send it to you. I build AI systems like this for service businesses: voice agents, automations, internal tools. Book a call: https://thisisaqib.com Chapters: 00:00 The cold open 00:20 Why I built this 00:32 The whole pipeline, three stages 01:24 Trigger and the settings tab 02:02 Reading the CV 02:17 The bug that multiplied everything by seven 02:42 Finding jobs: RSS boards 04:13 Finding jobs: LinkedIn and the concurrency limit 05:17 Merging sources and hashing for dedup 06:09 Moving dedup in front of the loop 06:42 Scoring each job against the CV 07:55 Draft, skip, or log either way 09:03 Logging and alerting Slack 09:30 Recap and the giveaway