Screen 50 Resumes At Once: Score, Fix, and Rebuild It Against the Job (Free, On-Device)

Batch Score & Triage in One Local Pass

Drop the 50 PDFs into the tool’s input directory. Run the score_batch.py script (or click “Analyze Folder” in the GUI) with the target job description (JD) pasted into jd.txt. The engine parses locally—no upload—extracting entities: hard skills (Python, Kubernetes, GAAP), soft-skill signals (“led,” “negotiated”), experience density (years × scope), and education/cert matches. It outputs a scores.csv with columns: filename, hard_match_pct, soft_signal_score, exp_years, gap_list, composite_score.

Immediately filter: keep rows where composite_score >= 70 AND hard_match_pct >= 60. Flag 60–69 composites for “skills gap fix.” Reject < 60. Open the top 10–15 CSVs in a spreadsheet; sort by gap_list length ascending. Candidates with 1–2 missing keywords but high experience density are faster to fix than those with 5+ gaps. Export the shortlist filenames to shortlist.txt for the rebuild phase.

Auto-Fix Gaps & Rebuild Tailored Resumes

Run rebuild.py --input shortlist.txt --jd jd.txt --template ats_clean.docx. For each candidate, the script:
1. Maps missing JD keywords to the candidate’s actual experience using a local embedding model (e.g., sentence-transformers/all-MiniLM-L6-v2) to find semantic equivalents (e.g., JD wants “CI/CD,” resume has “Jenkins pipelines” → rewrite bullet to “Built CI/CD pipelines with Jenkins”).
2. Injects the top 3 missing hard skills into the Professional Summary and Core Competencies sections.
3. Rewrites the 3 most recent role bullets using the STAR pattern (Situation, Task, Action, Result) pulled from the raw text, forcing quantified metrics (%, $, time saved).
4. Strips columns, graphics, headers/footers, and non-standard fonts; outputs a single-column, ATS-safe .docx named FINAL_[Name]_[Role].docx.

QA spot-check: open 3 random outputs. Verify no hallucinated metrics (script tags uncertain numbers with [VERIFY]). Replace bracketed placeholders manually. Batch-convert final