Job Search Autopilot

A personal job search automation tool — and a practical introduction to vibe coding.

This project does two things. It runs every morning, searches 13+ job sources, rates every listing with Claude AI, and emails you a ranked digest. And it's designed so that anyone — developer or not — can pick it up, describe what they want to an AI, and make it their own without writing code from scratch.

If you're job hunting, it saves you hours of manual searching. If you're new to AI-assisted development, it's a real working project you can learn on.

Screenshots

Dashboard

Health tab

Skipped jobs

Example email digest

Console output

What it does

• Searches 13+ sources: ATS direct (Greenhouse/Lever/Ashby), Workday, Adzuna, LinkedIn, Brave, Tavily, WeWorkRemotely, Jobicy, Himalayas, USAJobs, UltiPro/UKG, RemoteOK
• Filters out non-US roles, onsite roles outside your metro, below-salary-floor postings, staffing agencies, expired postings, and noise pages
• Extracts salary ranges from job descriptions when not provided as structured data
• Rates each job with Claude AI (Excellent Fit / Good Fit / Worth a Look / Skip) based on your profile
• Emails a daily digest sorted by tier
• Tracks your pipeline on a Kanban dashboard (Reviewing → Applied → Interviewing)
• Dashboard Health tab shows run history, source performance, and filter drilldown — click any filter pill to see which jobs were caught by it
• Skipped tab shows plain-English explanations of why each job was filtered

Making it yours — vibe coding with AI

This tool is built to be customized. The filters, search queries, company list, and rating logic are all plain text or simple Python lists — readable and editable even if you are not a developer.

The fastest way to personalize it is to upload the files directly into Claude or ChatGPT and describe what you want in plain English. You do not need to understand every line of code. Just tell the AI what to change and it handles the syntax. This is vibe coding — you drive with plain English, the AI does the editing.

How to do it

1. Go to claude.ai or chatgpt.com
2. Start a new conversation
3. Upload both of these files as attachments: config.example.py and job_autopilot.py
4. Paste the prompt below with your details filled in
5. Save the AI's response as scripts/config.py
6. Run it: python scripts/job_autopilot.py --run

The one prompt that sets up everything:

I've attached config.example.py and job_autopilot.py from a job search
automation tool. I need you to fill in config.example.py for my situation
so I can use it as my personal config.py file.

Here's my info:
  Name: [Your Name]
  Current role: [Your Title] at [Your Company]
  Target roles: [e.g. Software Engineer, Data Scientist, Marketing Manager]
  Target industries: [e.g. healthcare tech, SaaS, fintech]
  Target salary: $[amount]+
  Location: [e.g. "Remote nationwide" OR "Remote or hybrid in Chicago area"]
  My city and nearby suburbs: [list them]
  Things I never want: [e.g. contract roles, crypto, people management]
  My core strengths: [list 3-5]

Please fill in ALL sections of config.example.py and return the complete
file so I can save it directly as my config.py.

You can do multiple rounds — start with the basics and iterate from there. The AI can also help you debug errors, add new features, or explain what any part of the code does.

Debugging tip: If something breaks, paste the full error message plus the last 20 lines of console output into Claude or ChatGPT. You do not need to understand the error yourself — just describe what you were trying to do when it happened.

More prompts for fine-tuning

Set up for your role and location:

I've attached job_autopilot.py and config.example.py. I'm a [your role] looking
for [target roles] in the [your city] area. Can you:
- Update TARGET_TITLES in config.py for my role type
- Set LOCAL_METRO_TERMS to my metro (city and nearby suburbs: [list them])
- Fill in all the search query lists with my actual target titles and industry
- Rewrite the PROFILE block for my background: [brief summary]
- Rewrite the Claude rating prompt tier definitions for my situation

Add hard filters:

I want to automatically skip jobs that mention [thing you hate], [industry
you don't want], and any contract positions. Can you add these to
HARD_DISQUALIFIERS in config.py and explain what each one does?

Build your company watchlist:

Can you help me find the Greenhouse, Lever, Ashby, and Workday ATS slugs for
these companies and add them to the COMPANIES and WORKDAY_COMPANIES lists in
config.example.py: [list]

Tune the rating prompt:

The Claude rating prompt in job_autopilot.py is generic. I'm a [your role].
Can you rewrite the Excellent Fit, Good Fit, Worth a Look, and Skip tier
definitions for my situation? My target roles are [roles]. My core
strengths are [strengths]. Things I won't do: [list].

Debug a crash:

I ran python scripts/job_autopilot.py --run and got this error:
[paste the full error message]

Here is the last 20 lines of console output:
[paste console output]

What is wrong and how do I fix it?

Debug quiet days (no jobs in the email):

The autopilot ran and sent a quiet day email — 0 jobs made it through. Here
is the filter breakdown from the console output:
[paste the filter breakdown lines]

My target roles are [your roles]. My location is remote nationwide OR
on-site in [your city]. What is likely causing everything to be filtered?

The SETUP.txt file has more detailed prompts for each specific part.

What you can customize

In config.py

TARGET_TITLES — the job title keywords used to filter results from ATS boards, Workday, Jobicy, RemoteOK, and UltiPro. A job must contain at least one of these keywords in its title to be kept. Update this list to match your target role types.

TITLE_BLOCK_WORDS — words that disqualify a title even if it matched TARGET_TITLES. Use this to drop wrong-function roles (e.g. "engineer", "designer") or wrong-seniority roles (e.g. "intern") that sneak through.

PROFILE — the most important thing to personalize. Claude reads this when rating every single job. Write it like a concise professional summary: your current role, your background, what you are targeting, your strengths, and your hard constraints (location, salary, types of work you will not do). The more specific you are, the better the ratings.

MIN_SALARY — your salary floor as a plain number (e.g. 120000 for $120K). Jobs with a confirmed range entirely below this are filtered before Claude sees them. Jobs with no salary listed always pass through.

LOCAL_METRO_TERMS — your city and surrounding suburb names. The autopilot uses this to identify local hybrid and onsite roles and show them in a separate section of your email.

COMPANIES — the direct ATS company watchlist for Greenhouse, Lever, and Ashby. The autopilot checks all three for every company in this list on every run.

WORKDAY_COMPANIES — companies that use Workday ATS. Format: (tenant, wd_server, site, display_name). Find a company's Workday URL on their careers page — it follows the pattern https://{tenant}.{wdN}.myworkdayjobs.com/en-US/{site}.

VACATION_START / VACATION_END — set both to past dates if you are not on vacation. During vacation the autopilot buffers results and sends one digest when you return.

Search queries (ADZUNA_QUERIES, BRAVE_QUERIES, TAVILY_QUERIES, LI_REMOTE_QUERIES, LI_LOCAL_QUERIES, HIMALAYAS_QUERIES, JOBICY_QUERIES) — tailor these to your exact job titles and industry keywords.

HARD_DISQUALIFIERS — phrases that auto-skip a job before it reaches Claude. Use this for things you are 100% certain you never want.

HARD_DISQ_PATTERN — a regex string for disqualifiers that need word-boundary matching (e.g. r"\bdefi\b|\bcrypto\b").

COMPANY_PREFILTER — staffing agencies and job aggregators to block by company name.

WRONG_TITLE_PATTERNS — regex patterns that filter out wrong-function job titles before Claude sees them.

DOMAIN_COMPANY_MAP — maps URL domain substrings to company display names. When Brave or Tavily finds a job on a Workday page (e.g. capitalone.wd12.myworkdayjobs.com), this map tells the autopilot the company is "Capital One" instead of showing a blank or generic name. Add "{tenant}.wd": "Company Name" for each entry in your WORKDAY_COMPANIES list.

JUNK_URL_PATTERNS — regex patterns for URLs that are generic hiring portals or landing pages (not actual job postings). Jobs matching these are auto-skipped before Claude sees them.

In job_autopilot.py

SALARY_FLOOR_EXEMPT — companies you know pay well above your floor even when a search snippet shows a misleadingly low number.

_URL_CITY_RE and _LOC_CITY_RE — compiled from URL_CITY_PATTERN and LOC_CITY_PATTERN in config.py. Edit those config variables to control which cities are blocked for onsite roles.

Claude rating prompt — the tier definitions and rules Claude uses to score every job. Find the section starting with "Perfect Fit" = 90-100% in job_autopilot.py.

Quick Start

New to this? Read SETUP.txt for step-by-step instructions and troubleshooting tips. The steps below are the short version.

0. Download the project files

If you're on this page on GitHub and don't have the files yet:

1. Click the green Code button near the top of this page
2. Click Download ZIP
3. Unzip the downloaded file — you'll get a folder called Job-Search-Autopilot-main (or similar)
4. Rename it to Job-Search-Autopilot if you like, and move it somewhere easy to find (like your Desktop or Documents folder)

All the steps below are run from inside that folder.

If you're comfortable with Git, you can also run git clone https://github.com/crchalfant/Job-Search-Autopilot.git instead of downloading the ZIP.

1. Install Python (if you haven't already)

You need Python 3.10 or later. Download it at python.org/downloads. During installation on Windows, check the box that says "Add Python to PATH".

2. Install dependencies

Open a terminal inside your project folder:

• Windows: Open the folder in File Explorer, click the address bar at the top, type cmd, and press Enter
• Mac: Right-click the folder in Finder → "New Terminal at Folder"

Then run:

pip install -r requirements.txt

If you get a "pip not found" error, try: python -m pip install -r requirements.txt

3. Set up API keys

Windows:

copy .env.example .env

Mac/Linux:

cp .env.example .env

Then open .env in a text editor (on Windows: right-click the file → Open With → Notepad) and fill in your API keys. See the API keys needed section below for where to get each one.

4. Set up your config

Windows:

copy config.example.py scripts\config.py

Mac/Linux:

cp config.example.py scripts/config.py

Then open scripts/config.py in a text editor and customize it for your job search. Use the AI prompts in the Making it yours section above to fill it in quickly — you can have Claude or ChatGPT do most of the work.

5. Run it

python scripts/job_autopilot.py --run

The first run takes 5–10 minutes. When it finishes, check your email inbox for the digest.

Add --verbose to see source latencies and filter breakdown. Full setup instructions are in SETUP.txt.

Scheduling

macOS/Linux — add a cron job:

0 7 * * 1-5 cd /path/to/Job-Search-Autopilot && python scripts/job_autopilot.py --run

Windows — use Task Scheduler:

• Program: python
• Arguments: scripts\job_autopilot.py --run
• Start in: C:\path\to\Job-Search-Autopilot

Running the dashboard

python scripts/dashboard.py

Then open http://localhost:5000 in your browser.

Important: Keep the terminal window open the entire time you use the dashboard. The dashboard stops working if you close it. To stop the dashboard, press Ctrl+C in the terminal.

API keys needed

Getting all the keys takes about 15 minutes.

Anthropic (Claude AI) — required

Used to rate every job. The only paid service — costs roughly $1–5/month depending on run frequency.

1. Go to console.anthropic.com
2. Create an account and add a payment method
3. Go to API Keys → Create Key
4. Copy the key — starts with sk-ant-

Brave Search — required

1. Go to api.search.brave.com
2. Create a free account → API Keys → Create API Key → select Free plan
3. Copy the key — starts with BSA

Tavily — required

1. Go to app.tavily.com
2. Sign up for a free account — key shown on dashboard immediately
3. Copy the key — starts with tvly-

Adzuna — required

1. Go to developer.adzuna.com
2. Register → Dashboard → API Access Details
3. Copy both your App ID and App Key

Gmail App Password — required

1. Enable 2-Step Verification at myaccount.google.com/security
2. Go to myaccount.google.com/apppasswords
3. Create an app password named Job Search Autopilot
4. Copy the 16-character password

USAJobs — optional

1. Go to developer.usajobs.gov → Request an API Key
2. Set both USAJOBS_API_KEY and USAJOBS_EMAIL

Adding companies to the ATS list

Greenhouse / Lever / Ashby

The COMPANIES list in config.py checks all three APIs automatically for each slug.

ATS	URL pattern	Slug
Greenhouse	boards.greenhouse.io/stripe	stripe
Lever	jobs.lever.co/plaid	plaid
Ashby	jobs.ashbyhq.com/mercury	mercury

Workday

The WORKDAY_COMPANIES list in config.py covers companies that use Workday ATS. Format: (tenant, wd_server, site, display_name). Find a company's Workday URL on their careers page — it follows the pattern https://{tenant}.{wdN}.myworkdayjobs.com/en-US/{site}.

Project structure

Job-Search-Autopilot/
├── scripts/
│   ├── job_autopilot.py     ← main autopilot (run this daily)
│   ├── dashboard.py         ← Flask Kanban dashboard
│   ├── autopilot_shared.py  ← shared utilities (IDs, normalization, validation)
│   └── config.py            ← your personal config (gitignored)
├── config.example.py        ← copy to scripts/config.py and fill in
├── .env.example             ← copy to .env and fill in API keys
├── requirements.txt
├── SETUP.txt                ← full setup guide with customization prompts
├── README.md
├── LICENSE
└── output/
    └── Job Search Autopilot/  ← runtime files (gitignored)

Output files

All runtime files live in output/Job Search Autopilot/ (gitignored).

File	Purpose
*-jobs.json	Rated jobs for the dashboard board tab
*-skipped.json	Skipped jobs for the dashboard skipped tab
*-report.md	Email report attachment
.seen.json	Deduplication history (30-day rolling window)
board_state.json	Dashboard card positions and notes
autopilot_runs.db	SQLite run history for the Health tab
debug_job_log.txt	Full job log for debugging — upload to Claude to verify data accuracy
vacation_buffer.json	Jobs buffered during vacation — cleared automatically on return day

Vacation mode

Set VACATION_START and VACATION_END in config.py to your travel dates. The autopilot buffers results daily and sends one combined digest on your return day.

License

MIT + Commons Clause. Free to use for personal job searching. Reach out before building anything commercial on top of it. See LICENSE for details.