ScreenEval

ScreenEval is an internal tool Sam built for our recruiting team. Paste in the transcript of a recruiter's phone screen and it evaluates the candidate, coaches the recruiter on how the screen went, and adds the call to a searchable archive. Sam built it in about six hours of work between meetings, Monday afternoon to Wednesday, with Claude Code and Supabase. He showed it in Episode 29 (January 31, 2026), when it was about a day from going live.

Why we built it

Our recruiters run dozens of 20 to 25-minute screens every day. Everyone uses AI note-takers now, but those summaries stay with the person who took them. Sam wanted three things: consistent screens, coaching for recruiters, and hiring data the whole team can search.

We have a rubric of about six areas a recruiter should cover in a screen, and it's hard to hit them all in 20 minutes. Did they dig into past experience? Discuss compensation? Cover next steps?

The stack

• Claude Code in the terminal: built the app. Sam started briefly in Claude Cowork but found the terminal made more sense.
• Supabase: the database. Sam finds it "very easy to understand."
• Claude Opus 4.5: all the analysis. Sam at 10:29: "Opus 4.5 for everything."
• AWS Amplify and GitHub: our standard deploy, to a Ryz Labs subdomain.

How we built it, step by step

1. Build in Claude Code, store in Supabase. At 07:34 Sam summed it up: "But this was built in Cloud Code, Supabase, and that's pretty much it."
2. Paste a transcript. Drop in the AI note-taker transcript, plus the job description if you have one. The app fills in candidate, role, recruiter and date.
3. Record the recruiter's call first. The recruiter submits their own assessment before seeing the AI's, and isn't expected to change it. Recruiters catch red flags the AI misses.
4. Evaluate the candidate. Opus 4.5 scores communication, experience and enthusiasm, with key evidence.
5. Evaluate the screen. Rapport, follow-up probing, compensation, timeline and next steps, pacing, plus suggested questions to ask next time.
6. Make transcripts searchable. For example, every candidate who mentioned a certain former employer, or Python.
7. Add a manager dashboard. Thirty days of screens become specific coaching per recruiter.
8. Deploy on Amplify and GitHub.

How it turned out

In the demo, Sam's call and the AI's matched: move forward. His own coaching report said he was weak at probing follow-up questions about experience and could sell the opportunity better. Jordan spotted a likely bug in the "similar candidates" list but called the demo incredible.

But it's pretty incredible we could build this or I could build this with who I haven't written one line of code in my life, you know, full stack web app development in six hours that has some pretty cool features.

— Sam Nadler, Episode 29

It wasn't done. Sam expected bugs once recruiters used it; a similar tool he shipped that Monday had already needed three or four fixes.

What we'd do differently

• Let AI reinforce, not override. The recruiter's decision stands. The AI is there for coaching and a second opinion.
• Go real time. Jordan's suggested next step: replace Granola with our own tool and coach recruiters during the call, not after.
• Automate the input. Pasting transcripts by hand should go away.

Later that episode, Sam started a new app live in Cowork, the Hollywood transformation trainer. See Claude Code vs Claude Cowork and our other recruiting tools, Ryz Score and NTRVSTA.

FAQ

How do you build an AI recruiter screen evaluation tool?

Sam built ScreenEval with Claude Code in the terminal and Supabase as the database, using Claude Opus 4.5 for all the analysis. You paste a screen transcript from an AI note-taker, the recruiter records their own call, and the app returns a candidate evaluation and a coaching report on the screen itself.

How does AI coaching for recruiters work?

ScreenEval scores each screen on areas like rapport, follow-up probing, compensation, timeline and next steps, and pacing, then suggests follow-up questions. A manager dashboard rolls 30 days of screens into per-recruiter coaching.

Can you search interview transcripts with AI?

Yes. Once screens go through ScreenEval, recruiters can search all transcripts, for example for every candidate who mentioned a specific former employer or Python.

How long does it take a non-developer to build a tool like this with Claude Code?

Sam, who has never written a line of code, built it in about six hours of work between meetings, from Monday afternoon to Wednesday.