Michael Manapat

Michael Manapat is the co-founder and CEO of Rowspace, which builds AI tools for institutional investors. His work has moved from mathematical models of evolution to machine learning for payment fraud and, now, to organizing the accumulated knowledge financial firms use to make decisions.

From mathematics to production machine learning

Manapat earned a mathematics PhD at MIT. His research in evolutionary dynamics included models of trust: one result suggested that reputation could make transactions efficient while leaving trustees with most of the gains. The thesis also explored how systems that generate diverse sequences without replication might give rise to replicating life. He was a postdoctoral fellow in applied mathematics at Harvard and a software engineer at Google before joining Stripe.

At Stripe, Manapat managed engineering for Radar and machine learning, then led applied machine learning across fraud, credit, and forecasting. In his explanation of Radar, he described how its models drew on payment patterns across Stripe’s network and fraud information from banking partners. In 2018, Radar for Fraud Teams added controls that let risk teams set thresholds and test custom rules against historical transactions.

A challenge in fraud detection is that a model changes the data available to evaluate it: when a payment is blocked, its outcome cannot be observed. Manapat described how Stripe’s team adapted counterfactual evaluation to this problem. They allowed a small, randomly selected share of payments that would otherwise have been blocked, then weighted the observed outcomes to estimate what would have happened without blocking. If one in 20 payments in a group was allowed through, for example, each observed outcome could stand in for 20 payments in the estimate. The resulting data could help evaluate existing decisions and train new models.

The method came with costs and limits. Sampling more often near the decision threshold could produce information where the model was less certain while exposing fewer obviously fraudulent payments. More sampling improved estimates but could also incur more fraud; rare observations given very large weights could make estimates unstable. Manapat also noted that related attempts in a payment session may need to be treated together, because they are not independent events.

From Notion to institutional knowledge

Manapat later joined Notion, where he was vice president of engineering and then CTO, helping scale engineering and advance the company’s AI work. He subsequently co-founded Rowspace with Yibo Ling.

At Rowspace, Manapat is working on the problem of making financial firms’ scattered records usable in analysis. The company’s approach, described by its investor Emergence Capital, starts by reconciling data across systems, turning historical memos into machine-readable information, and mapping how a firm resolves conflicting evidence. Search and analytical agents can then draw on that foundation for work such as deal assessment and portfolio monitoring. The aim is to make an experienced investor’s judgment and a firm’s history useful to colleagues facing new decisions.