Grownout

Overview

An AI-powered hiring platform co-founded, architected, and scaled from the first line of code to strategic acquisition in under four years — matching candidates to roles using classical ML at a time when that meant being precise about what a model could and could not decide.

The problem

Hiring runs on judgement that is hard to encode and easy to get wrong. A ranking model that quietly encodes the wrong signal does real damage to real people's careers, and it does it invisibly. The platform had to improve on human shortlisting where the evidence supported it, and stay out of the way where it did not.

What we built

Candidate matching models

Classical ML ranking applied where it measurably outperformed heuristic shortlisting, and deliberately not beyond that.

Referral and network graph

The relationship data that made matches defensible, since a ranked list without a reason is not usable by a recruiter.

Employer and candidate products

Two surfaces with genuinely different incentives, built so improving one did not quietly degrade the other.

Team and platform

The engineering organisation built alongside the product, and the architecture that let a small team keep shipping.

How we approached it

• Be explicit about what the model decides and what it only suggests
• A ranking is only useful if the reason behind it can be shown
• Classical ML where it wins; heuristics where it does not
• Architect for a small team's throughput, because that was the actual constraint

Outcome

A hiring platform taken from first commit to strategic acquisition in under four years, with ML in production and an engineering organisation that outlasted the exit.