Most hiring pipelines still rely on manual resume review or brittle keyword matching, which makes fairness and reproducibility hard to maintain. This project turns resume PDFs into structured JSON via sectioned LLM prompts, augments the result with GitHub profile and repository signals, then applies a strict, template-driven evaluator that produces transparent scores and evidence.
What Sets It Apart
- Sectioned, template-first extraction: uses Jinja LLM templates per resume section so the output maps to a JSON Resume–style schema, improving consistency across providers and inputs — so what: downstream scoring receives normalized, auditable fields instead of freeform text.
- GitHub enrichment and project selection: fetches profile/repos, classifies projects, and forces an LLM-backed selection of a fixed number of repos with contribution thresholds — so what: technical signal is quantified and limited to meaningful contributions, reducing noise from minor forks or one-off repos.
- Fairness-aware evaluator: scoring templates encode category rules, bonus/deduction logic, and evidence explanations — so what: evaluations are reproducible and easier to inspect or adjust for bias concerns.
- Local-first operation: supports Ollama for offline runs or Google Gemini for hosted usage — so what: teams can run evaluations without sending resumes to third-party endpoints if desired.
Who it's for and tradeoffs
Great fit if you need an auditable, reproducible way to convert resumes into structured data and produce transparent candidate scores for triage or analysis. It helps engineering hiring teams who want to incorporate open-source activity as a quantified signal. Look elsewhere if you need turnkey ATS integrations, large-scale enterprise deployment, or a polished UI out of the box — this repo is an opinionated pipeline and requires engineering work to adapt scoring rules and integrate into production systems.
Where it fits
Acts as a middle layer in hiring workflows: PDF ingestion → LLM extraction → GitHub enrichment → template-based scoring → CSV/JSON export. Best used for pilot evaluations, fairness audits, or as a prototype component in a larger hiring platform.
