How Stitch's custom model learns your company

Finding candidates is only half the job. Once you have them, you have to decide who is actually worth interviewing, and that is where most tools fall back on generic filters. Stitch does something different: it scores every candidate with a model trained on your company. Here is how that works.

The three things that deliver talent

Stitch delivers world-class talent with three things working together. Data, so we can find the right candidate for any role you define. Scoring, so we surface only the best of them. And outreach, so they actually take the call. This post is about the middle one, the scoring, because it is what turns a billion profiles into a short list you would genuinely interview.

Scoring against your bar

For each role, the Stitch Swarm searches the live internet and gathers candidates, and the model scores every one against your specific role, surfacing only the top 0.1%. The funnel is steep on purpose:

• 1 billion+ profiles reachable across the internet.
• ~200,000 gathered and deeply scored for your role.
• 100 to 200 surfaced at your bar of excellence.

The scoring engine was built by our founding team, ML engineers from Microsoft and Google with 17 AI patents between us, in collaboration with world-class researchers. It is not a relevance score with your keywords pasted on top.

It's custom per company

The important part is that the model is yours. It is trained on your roles, your past decisions, and your definition of great, not on a global average. Two companies hiring for the same job title get different candidates from Stitch, because they have different bars, different cultures, and different ideas of what excellent looks like. A generic filter cannot do that, because it is the same logic for everyone.

How it learns

Your model learns from every candidate it considers, not just the ones you act on. The same model works through all 200,000 profiles one by one, updating itself as it goes, which is why scoring alone takes five to ten hours. Each profile teaches it more about the landscape of talent that actually exists, the shapes of real careers, and the signals that separate strong from exceptional. The more it sees, the better it understands what genuinely top-tier looks like, rather than what a job description claims it should.

It also learns from you, continuously: from your scoring reviews, your declined meetings, your interview transcripts, and your post-interview feedback. Every signal you give it teaches it more about what great actually looks like for you. The model is sharper after your hundredth decision than it was after your first, and it is yours alone. Nobody else searching the same internet gets your model.

That is the core difference between a custom model and a shared one, which we compare directly in Stitch vs Eightfold AI. Or start a 14-day trial and calibrate the model on your own roles.