HR Batch Resume Screening

What It Solves

When screening large resume batches, reviewers often open each PDF, compare it against the job description, and judge hard requirements, role fit, achievements, skills, and stability. That process is slow and hard to keep consistent. This skill targets locally searchable PDF resumes and provides a command-line script, screen.py, to run structured shortlisting with an LLM against a job description while preserving auditable scores and an interactive report.

How It Works and Limits

The core flow reads .pdf files from a folder, de-duplicates filenames in lowercase, extracts resume text, and scores each resume on:
- Basic hard-requirement fit: degree, years of experience, mandatory requirements
- Role responsibility fit: coverage of the JD responsibilities
- Achievement fit: project outcomes and measurable impact
- Technical skill fit: relevant stack, tools, and methods
- Stability and potential: tenure, career trajectory, and growth signal

Outputs include:
- screening_result.json: structured scores for downstream filtering and analysis
- screening_report.html: an interactive shortlist report with radar charts, summary tags, and expandable details

The skill only handles locally searchable PDFs; Word, Excel, PPT, images, and scanned PDFs must first be converted into searchable PDFs. If PDF text extraction fails, it falls back across pdfplumber, PyPDF2, and pdfminer.six; failed LLM calls are retried automatically. Write hard requirements and priorities explicitly in the JD, keep resume plus JD around 12k–14k characters where possible, and treat the composite score as a first-pass signal rather than the final decision.