Job Hunt — Setup

One-time profile generation from a folder of resumes.

What it does

Reads every resume in inputDir, sends the combined text to Ollama with a strict JSON-output prompt, and writes the resulting skills profile to $HOME/jobhunt/profile.json. The jobhunt-daily workflow reads this file to know what jobs to search for, how to score them, and how to tailor outputs.

Setup before first run

1. Put your resumes in /home/me/resumes (or edit inputDir in the YAML). Supported formats: .doc, .docx, .odt, .pdf, .rtf, .txt.
2. Create an Ollama credential named ollama pointing at your local Ollama server. Default endpoint is http://localhost:11434.
3. Pick a model. The YAML defaults to llama3.2. For higher-quality structured extraction, use llama3.1 (8B), mistral, or qwen2.5. Pull the model first: ollama pull llama3.2.
4. Make sure /home/me/jobhunt/ exists (or change the outputFile path in the YAML).

Don't have Ollama? Use Claude instead

Change two lines in the YAML:

  - name: extract-profile
    tool: anthropic                    # was: ollama
    credential: claude-api             # was: ollama
    properties:
      model: "claude-sonnet-4-6"       # was: "llama3.2"
      …

Tradeoff: Anthropic produces cleaner JSON on the first try but the resume content (incl. your full name, employment history, salary expectations) leaves your machine for every run. For one-time setup this is usually fine. For frequent re-runs, Ollama keeps it local.

Output

Single file: /home/me/jobhunt/profile.json (path is whatever you set as outputFile). Shape:

{
  "name": "Jane Doe",
  "currentRole": "Staff Software Engineer",
  "experienceYears": 12,
  "level": "staff",
  "summary": "Jane is a backend-leaning generalist with deep AWS and Postgres experience…",
  "skills": ["distributed systems", "API design", "team leadership", …],
  "languages": ["Go", "Python", "TypeScript"],
  "frameworks": ["Kubernetes", "AWS", "PostgreSQL", "React"],
  "domains": ["fintech", "developer tools"],
  "education": "BS Computer Science, Carnegie Mellon, 2012",
  "preferences": {
    "remote": true,
    "locations": ["Remote", "San Francisco"],
    "salaryFloor": 180000,
    "roleType": "ic"
  }
}

When to re-run

Whenever you update any resume in the folder. The output is deterministic for a given input — safe to overwrite each run.