About the company Our mission is that “Any financial application can onboard any user, anywhere in the world, in 1 click.” Transak provides onboarding to financial applications through authentication, KYC, risk checks, and fiat on/off ramps. This is a next generation of infrastructure for the next generation of financial applications that are built on blockchain and stablecoin rails. Our API and widget-based solutions are used by top partners like MetaMask, Coinbase, Ledger, and Trust Wallet to enable seamless onboarding of over 10 million users across over 450 active applications. We have raised over $37M from top-tier investors including Consensys, Tether, and Animoca Brands About the Role The mandate is easy to state and hard to deliver: stop fraud while approving as many legitimate transactions as possible. How you do it is yours to decide — deterministic heuristics, machine learning, AI agents, or whatever the problem demands — and the adversaries on the other side are among the most sophisticated in the world. You'll take genuine end-to-end ownership of how Transak detects and prevents fraud, as part of a small, senior team. The commercial stakes are direct: every basis point of fraud, and every unit of unnecessary friction, maps straight to revenue and to real users who can or cannot transact. The problem at scale This is a genuinely hard applied problem, and the conditions to do something exceptional with it are already in place. Tens of millions of transactions of history to learn from, with 60+ fields each. ~1,000 live risk signals available per decision, across 13 providers. ~10,000 orders per hour at peak, with value well into seven figures in a single peak hour. Tens of thousands of attempts in a single coordinated attack, over a matter of days. The adversary is intelligent and adaptive. This is not a static classification task — the distribution shifts the moment you respond, because the counterparty is actively working to defeat you. The modelling layer is largely greenfield. The data and the scale already exist; modern machine learning has not yet been applied to them properly. The opportunity to do so — and to own it — is wide open. Key responsibilities Own fraud, chargeback and transaction-risk models end to end — from framing the question, to features and rules, to what ships, to the thresholds that set policy. Build and tune the machine-learning and signal layer that operates alongside our external risk vendors. Set the direction of where risk modelling goes next, and bring the team with you. Act as a data scientist for the wider business — making data accessible through the Snowflake warehouse, self-serve tooling, dashboards and internal data assistants that let any team query our data directly. Take on high-leverage product, growth and experimentation problems as they arise. Help advance how we apply AI internally, from agentic coding assistants to internal copilots and novel uses of large language models. What we're looking for We weight how you think far more heavily than any checklist of tools. Strong mathematical and statistical foundations, and a genuine pull toward hard, quantitative problems. A self-starter who identifies the important problem, scopes it, and acts without waiting to be directed. A capable engineer and analyst — fluent in Python and SQL, and able to take a model from idea to something that runs and ships. Intellectually honest about uncertainty, and rigorous in evaluating your own work. A genuine interest in crypto, payments, fraud and the data they generate. You might be an exceptional recent graduate, a PhD, or a year or two into your career and ready for far more ownership than your current role allows. The non-negotiables are raw ability and drive. Desirable None of the following are required, but any would strengthen an application. A track record of exceptional output — a strong degree from a leading university, research, competition results (Kaggle, olympiads), open-source contributions, or something you built that people use. Exposure to fraud, risk, payments, crypto, or other adversarial, high-stakes machine learning. Experience deploying and maintaining models in production, including monitoring and drift. Depth in experimentation or causal inference. A habit of building your own tools and automations. Working with us Autonomy, against a clear results bar. You'll have the freedom to try new models, tools and approaches; we judge outcomes, not process or hours. AI-native by default. We use Claude, Codex and whatever is current, heavily, across engineering and analysis — and expect the same of you. Leverage is the point. A small, technical, high-trust team. Short feedback loops, little bureaucracy, and colleagues who hold a high bar.