In an ecosystem increasingly dominated by expensive, opaque AI agents, a new open-source project is proving that sometimes the best tool is just good old-fashioned code. Released on GitHub by developer daliparthi, the 'personal-job-agent' is a local Workday job application tool that relies entirely on rule-based logic for resume tailoring and application automation. It runs on Windows 10/11, requires no API keys, and promises zero per-resume costs, marking a distinct shift away from the LLM-heavy trend in job search automation.

Pure Code, No Hallucinations

The agent operates by searching 109 large employers' Workday career sites using public APIs, filtering by strict criteria like job type, salary, and location. The core differentiator is its tailoring engine: it parses the user's master resume into a YAML file and reorders existing content based on keyword matches from the job description. It explicitly avoids AI rewriting, stating that 'rules don't rewrite your bullets into new sentences.' Instead, it selects, orders, and re-words skill names to match ATS expectations, ensuring the output remains strictly grounded in the user's actual experience.

Local Privacy and Automation

Privacy is a central feature, with all data stored locally in directories like data/ and applications/. The dashboard listens only on 127.0.0.1, ensuring no personal data leaves the machine except for the actual application submission to Workday. The tool includes a browser automation component using Microsoft Edge or Playwright's Chromium to fill out application forms, including screening questions defined in a config file. It stops at the review page by default, allowing the user to verify details before clicking submit, mitigating the risk of automated errors.

Key Takeaways

  • The tool uses a built-in list of ~580 skills and synonyms to match job postings without AI interpretation.
  • It supports Windows Task Scheduler for daily automated runs at 11:00 AM Eastern.
  • Users must manually verify the parsed resume_data.yaml to ensure accuracy before applying.
  • The project is MIT licensed and provided 'as is,' with no warranty for application outcomes.

The Bottom Line

While AI agents are the current hype, this project reminds us that deterministic, rule-based systems offer transparency and cost-efficiency that LLMs cannot match. For developers who trust their own bullet points more than a model's creative rewriting, this is a solid, private alternative. It’s a refreshing hack for those tired of paying token fees for simple keyword matching.