An AI Matching Engine Built in 11 Weeks

gravity9 partnered with a Nordic-headquartered influencer marketing platform to solve a problem that was quietly capping its growth: turning a campaign brief into an influencer shortlist took up to a month of manual, relationship-driven work, and the process couldn’t scale with headcount. In just 11 weeks, gravity9 built an AI-based recommendation engine, running natively on MongoDB Atlas, that evaluates candidates against nearly 9,000 real past hiring outcomes rather than instinct or tribal knowledge. The five-layer system combines vector search, ten structural signals, a reranking model, and a learned ranking model trained on the client’s own hire history, then explains every recommendation in plain language so campaign managers can interrogate and defend each pick to their own clients.

The results speak for themselves: on 200 campaigns the model had never seen, the engine correctly recovers 67.88% of influencers who were actually hired, with a Mean Reciprocal Rank of 0.877, meaning the first genuine hire typically lands at or near the top of every shortlist. What once took up to a month now happens in minutes, and the client walked away with far more than they contracted for, including a full web application, live streaming results, and a transparent feedback loop that learns from rejected candidates, all while directly solving the “black box” concern they raised at the outset. As the client put it: “This is impressive, we are really happy with performance; such an improvement.”