Overview · Demo build
Calibrated AI screening that cites its work.
Drop in a job description and a stack of resumes. Get a ranked shortlist with strengths, gaps, and recommendation — each anchored to specific resume evidence. No signup.
- 0195Hire
Daniel Park
Confluent · Staff
- 0295Hire
Priya Iyer
Stripe · Staff
- 0365Phone screen
Sofia Costa
Cloudflare · Rust
- 0465Phone screen
Marcus Williams
Truist · Java
- 0530Pass
Asha Singh
Plaid · Junior
- 0630Pass
Jordan Kim
Shopify · Analyst
The brief
Recruiters waste hours on resumes the model can rank in twelve seconds.
Shortlist is a portfolio demo built on Next.js 16, Supabase, and Gemini 2.5 Flash. Background analysis via after() streams results to the dashboard while the API call returns instantly.
What buyers see
- Score with cited reasoning, not vibes
- Honest gaps — model is told not to flatter
- Calibrated rubric, not a Likert scale
- Structured output — JSON schema, no LLM parsing tricks
- Background work — POST returns in 80ms
Three ingredients
What makes the rankings worth reading.
Calibrated scoring
90+ reserved for genuinely exceptional candidates. Most stay below 80. Score creep is the death of AI screening.
Reasoning with evidence
Every score points to specific phrases from the resume. 'Strong communicator' is banned; 'led the Kafka migration at 40k tx/sec' is the standard.
Gaps are named bluntly
If a required skill is missing, the model says so. No 'transferable skills' theater.