Find roles faster
Pull new postings from the sources I use instead of checking each one by hand.
← All projects
Project · AI-assisted workflow
Building an AI job search system I can trust
A job search is a pipeline: find roles, judge fit, tailor materials, track follow-ups. I defined the requirements for career-ops, directed an AI build of it with Claude Code, and set one hard rule: no resume claim without evidence.
The problem
Searching well takes more steps than people admit
A serious search means watching many job boards, reading long postings, rewriting a resume for each one, and remembering who said what.
Generic AI resume tools make this worse in one way: they will invent experience to match a posting. I wanted speed without that risk.
Judge fit honestly
Score each role against my background so effort goes to the right ones.
Stay truthful
Tailor resumes without changing employers, titles, dates, or metrics.
What it does
One pipeline from posting to follow-up
I acted as product owner and Claude Code acted as the engineer. I wrote the requirements, reviewed the results, and decided what shipped.
Scan
Collects new postings from Greenhouse, Ashby, Lever, and LinkedIn, removes duplicates, and filters out locations and titles I do not want.
Score
Reads each full job description and rates fit against my profile, so the dashboard shows a ranked list instead of a pile of links.
Tailor
Builds a role-specific resume from my verified career evidence. It asks me about skill gaps before writing, and it will not add a claim I have not confirmed.
Check
Compares dollar amounts and percentages in the draft against my fact store. A number that is not in the evidence does not pass.
Track
Reads recruiter email, updates each application's status, and keeps interview notes in one dashboard.
What I learned
Customer success habits carried over
Building it felt like running an onboarding: define success, find where the process breaks, and measure whether it works.
Requirements decide quality
Vague instructions produced vague software. Specific rules, such as the evidence gate and a banned-phrase list, produced reliable behavior.
Match the model to the job
A small local model handled simple extraction. Resume writing needed a stronger model, so I moved that step to Claude only.
Test what matters
The project has an automated test suite so a change to one step does not quietly break another.
Automated test files
Job sources scanned: Greenhouse, Ashby, Lever, LinkedIn
Verified fact store behind every resume claim
What this project demonstrates
I can turn a messy, repeated process into a defined workflow, direct AI tools toward it, and put guardrails where accuracy matters. That is the same approach I bring to customer onboarding, adoption, and renewal work.
Built with Claude Code under my direction. I define requirements and review results; I am not a software engineer. The source code is private because it handles personal career data.