Our summary
A hiring team may receive more résumés than it can read carefully. Sharvin Shah builds a tool that sorts applications against criteria chosen for each job. It helps people decide what to read next, rather than rejecting applicants or deciding whom to hire.
The team writes questions and sets weights for the answers. Jev reads the résumé and returns ratings, labels and yes-or-no probabilities. Ordinary code calculates the total; labels remain separate because they have no numerical order. When criteria change, another screening is saved without erasing the old one.
The author reports costs and delays from one early run, not better hiring outcomes. A visible score can still bias a reviewer, and job criteria can unfairly reward opportunities some applicants never had. Scanned PDFs are rejected, and people must check uncertain answers and make the final decisions.
Key takeaways
- Write criteria for the specific job and let reviewers inspect how each answer affects the total.
- Keep arithmetic in code. A category label, such as a type of career progression, should not become a score just because it is an AI answer.
- Save each screening separately so a changed rubric does not erase the earlier judgment. Human review and fairness checks remain necessary.