Built This Week/Builds

Builds

NTRVSTA

NTRVSTA started as our internal fix for thousands of first-round interviews a year: an avatar interviewer that runs dynamic interviews 24/7, scores candidates and plugs into the ATS.

NTRVSTA is the AI interview platform we built at Ryz Labs. An avatar interviewer named Carolina runs first-round interviews at any hour and in many languages, asks follow-up questions based on each answer, and scores candidates on technical and soft skills. Recruiters get chaptered video playback, transcripts and per-question evaluations inside their ATS. We launched it externally in Episode 6 (August 1, 2025), the same day it went up on Product Hunt. Unlike most builds on this show, it was not vibe-coded in a few days: Sam said as much at 16:49, and it had been in development long enough to run thousands of interviews already.

Why we built it

We do thousands of interviews a year, each 30 to 60 minutes, often with candidates who aren't a fit. Our Ryz Score app had already cut resume screening time, but interviewing was still "the big time suck." Jordan at 04:09:

And so it started off with just, a really simple idea of, you know, how can we use AI and AI avatars to help us interview.

— Jordan Metzner, Episode 6

His case for an AI interviewer: unlimited availability and scale, less bias, consistent scoring, and expertise across many subjects. To find the best people you have to widen the top of the funnel, and AI is the only way to interview that many candidates.

The stack

  • ChatGPT and other AI tools: scoring, translation and feedback.
  • Text-to-speech, speech-to-text and avatar generation: Carolina's voice, hearing and face.
  • Video streaming and video editing: the live, meeting-like interview and the chaptered playback.
  • Lever: our ATS, where most of our recruiters see results.

Jordan didn't name specific vendors for the voice, avatar or video pieces.

How we built it, step by step

  1. Start from our own problem. It began as an internal project right after Ryz Score.
  2. Use LLMs for judgment. ChatGPT and other models score answers, translate and write feedback.
  3. Stitch together the media pipeline. Speech-to-text, text-to-speech, avatar generation, streaming and editing combine into one end-to-end interview that feels like a real meeting.
  4. Prompt-engineer and red-team the avatar. We worked hard to keep Carolina on topic, friendly and appropriate, and tried to make her say things she shouldn't. Jordan said results were really good.
  5. Build the assessment builder. Upload a job description or pull it from the ATS, add topics, questions and instructions, optionally turn on whiteboard mode, then edit the generated assessment or send it as is.
  6. Integrate with the ATS. NTRVSTA becomes a secondary view inside Lever, so recruiters keep their existing flow.
  7. Add the admin layer. User management, dashboards and analytics.

How it turned out

In the demo, a dermatologist interviewing for an RLHF role scored 89.5 overall (90 technical, 85 soft skills). The review screen shows a headshot, highlights and lowlights, chapters, multilingual subtitles, playback speed, every question with its evaluation, the full transcript, and analysis of a diagram she drew in whiteboard mode.

Candidates go from Ryz Score to an interview within minutes instead of waiting days to schedule. They adapt to the avatar quickly, and emailing them their score afterward builds buy-in. Sam noted the latency is not as fast as two humans talking, but it isn't awkward. The headline number came at 19:14:

You know, in in the last thirty days, we've we've increased our recruiting capacity, our recruiters by 25%.

— Sam Nadler, Episode 6

Instead of cutting recruiters, we hired more, because each one can take on more roles.

What we'd do differently

  • Integrate first. The ATS integration, not the standalone interface, is what worked best internally.
  • Keep humans later in the funnel. AI interviews add capacity at the top; human interviews still happen after.
  • Plan to keep upgrading. The AI tools improved a lot while we built it, and Jordan expected new models to keep improving it.

Months later, in Episode 36, Sam used videos and transcripts from our AI interview tool (listed as EntreVista) to build the candidate highlight reel builder. For other recruiting builds, see Sam's ATS and AI in Recruiting.

FAQ

What is NTRVSTA?

NTRVSTA is an AI interview platform where an avatar interviewer, Carolina, runs dynamic, multilingual interviews at any hour, asks follow-ups based on answers, and scores candidates on technical and soft skills. It started as an internal Ryz Labs project and launched on Product Hunt the day Episode 6 aired.

How do you build an AI interviewer for recruiting?

We combined ChatGPT and other AI tools for scoring, translation and feedback with text-to-speech, speech-to-text, avatar generation, video streaming and video editing, then did a lot of prompt engineering and red-teaming to keep the avatar on-topic and appropriate.

Does an AI interviewer replace recruiters?

Not for us. We still do human interviews later in the funnel. Since adopting NTRVSTA we have hired more recruiters, and Sam said recruiting capacity grew 25% in 30 days.

Does the AI interview tool integrate with Lever?

Yes. We use Lever as our ATS, and most of our team uses NTRVSTA through that integration. Jordan said it also integrates with other ATS and CRM tools such as Greenhouse, Ashby, Workday and Salesforce.