Hyperscreen
2025
HyperScreen
Project Manager · Amazon Capstone · Jan 2025 – Apr 2025 · Vancouver, BC
AWS Vancouver postings routinely draw 750+ applicants, and AI application tools like Simplify and AIApply have made that number a trend rather than a spike. HyperScreen is the screening platform we built for their recruiters and hiring managers: applicants apply through a public portal, hiring managers set weighted scoring criteria per posting, and the system ranks the pool so first-round screening starts from a shortlist instead of a stack. The target was a 50% cut in screening time without trading away hire quality.
What I owned
I ran the requirements and the release plan: the terms of reference, the business requirements document, the project plan, and the module-by-module schedule the seven engineers built against.
I assigned every task to two people. On a team where everyone has a full course-load, a member going dark during midterm season needs to be a risk mitigated from the start. Pairing every task made that survivable by design.
Scope
Four modules, each with its own design, dev and release dates:
| Module | What it does | Target release |
|---|---|---|
| Applicant | Public, no-auth job search and application with resume upload | Feb 17 |
| Hiring Manager | Create postings, set and prioritize scoring criteria, evaluate and rank | Feb 17 |
| Recruiter/Admin | Hiring-manager onboarding and accounts, master scoring criteria, launch evaluations | Mar 14 |
| Reporting | Applicant counts, rankings, per-applicant detail, exports and visualizations | Mar 14 |
Success metrics
I defined the metrics against manual screening:
- 75% of applicants scoring above 80 should also be green-lit by first-round manual screening.
- 50% of HyperScreen hires still employed after one year.
- 50% reduction in average screening time, with all user actions responding in under 2 seconds
Build
Django 5 with DRF over PostgreSQL 17, a Next.js 15 / React 19 frontend in TypeScript and Tailwind, and django-q2 running resume evaluation as background jobs. Production is four Docker containers behind nginx, with the frontend, backend and database on a shared internal network. Resume parsing ran through PyMuPDF and python-docx, and scoring ran on a local Ollama model to keep inference free.