The workflow before
Point72 had 10 years of recruiting history: resumes in PDF, Word and scanned form, search firm assessments, interview notes and hiring outcomes, spread across separate systems. None of it could be searched or compared at scale. Recruiters and analysts read documents one by one, each hiring team applied its own criteria, and the firm could not learn from which past hires had worked out.
What we built
We built an HR data lake on AWS that ingests documents from the applicant tracking system, HRIS and document stores, runs scanned files through Amazon Textract, and standardizes and links records for the same candidate across sources.
On top of it we built a recruiting application with search across all historical talent data, automatic extraction of qualifications and experience, and a ranking model. Topic models (pLSA and LDA) turn free text into comparable profiles, classifiers sort and clean documents, and a logistic regression model trained on past hiring outcomes scores candidates against benchmarks. SHAP values show recruiters why a candidate ranks where they do.
Reviewers validate extractions and label edge cases through Pybossa, and the models retrain as new hiring outcomes arrive. The work ran about 12 months, from architecture through phased rollout to recruiting teams.
Before and after
| Measure | Before | After |
|---|---|---|
| First-pass document review | Hours per candidate | Minutes per candidate |
| Historical talent data | Free text in separate systems | 10 years in one searchable store |
| Candidate ranking | Manual, varied by team | Scored against shared benchmarks, with reasons |
What's next
The same data lake and model pipeline can take on Level 3 work: an agent that screens new applications end to end, drafts a candidate summary and ranking, and sends only borderline cases to a recruiter.
