Machine Learning Engineer (Onchain Risk)

Saifuro checks payments that AI agents try to make. An agent asks to spend, we compare the request with a mandate a human approved earlier, and the answer comes back signed so the receiving side can verify it. There's escrow on Base and Solana for trades between agents, and settlement is in USDC, onchain. We never hold customer funds. The docs are public at docs.saifuro.com, and the people who use them are developers building agents and the teams selling APIs to those agents.

A mandate catches what it was written to catch: the limits, the allowed sellers and the budget. It misses an agent that stays inside all of them but acts unlike itself. Say it pays an allowed seller, but at an address that first appeared onchain the day before, or splits one large purchase into forty payments just under the limit. We want a model that can tell whether an agent's spending still looks like what its owner approved. You may also know the role as ML engineer, applied scientist or risk data scientist. You work directly with our CTO.

What you'd be doing

• Risk signals. Models that score each payment from the agent's own history and what the chain shows about the other side, fast enough to run inside a decision.
• Explanations. Every flag comes with a reason a finance team can read, since they're the ones approving or rejecting the payment.
• Feedback. Those approvals and rejections go back into the models. You measure both kinds of mistake: payments held for no reason, and payments that should have been held.
• Budget forecasting comes later in 2027: how much each agent will spend next month, learned from how it spent before. The model would be yours.

What we need from you

• You write production Python.
• You've shipped a model that made or shaped decisions about money or risk, in fraud, credit, abuse or anomaly detection, and watched how it did after launch.
• You live in one of the countries we can hire in.

Beyond these requirements, we'll support you in learning whatever else the role demands. If your experience is relevant but your title differs, we would still like to hear from you. Work outside a job counts too: open source, hackathons and your own projects.

It helps, but isn't required, if you've worked with onchain data, if you've used graph or sequence models on transactions, if you've served features in real time, if you know PyTorch, XGBoost or LightGBM, or if you've had to explain a model's decisions to people who could overrule it.

Countries we can hire in

The role is fully remote. You can live and work in any of the countries below.

• Americas. Argentina, the Bahamas, Brazil, Canada, Chile, Colombia, Costa Rica, Dominican Republic, Ecuador, Guatemala, Mexico, Panama, Paraguay, Peru, the United States (including Puerto Rico), Uruguay.
• Europe. Albania, Austria, Belgium, Bosnia and Herzegovina, Bulgaria, Croatia, Cyprus, the Czech Republic, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, Iceland, Ireland, Italy, Latvia, Liechtenstein, Lithuania, Luxembourg, Malta, Monaco, Montenegro, the Netherlands, North Macedonia, Norway, Poland, Portugal, Romania, Serbia, Slovakia, Slovenia, Spain, Sweden, Switzerland, Turkey, the United Kingdom (including Jersey, Guernsey, the Isle of Man, Gibraltar, Bermuda, the Cayman Islands and the British Virgin Islands).
• Middle East and Africa. Bahrain, Israel, Kuwait, Mauritius, Oman, Qatar, Saudi Arabia, South Africa, the United Arab Emirates.
• Asia-Pacific. Australia, Brunei, China (including Macau), Hong Kong, India, Indonesia, Japan, Malaysia, New Zealand, the Philippines, Singapore, South Korea, Taiwan, Thailand, Vietnam.

Pay and benefits

• Salary $205,000 to $255,000. That's the US range. Outside the US we pay in USD, or in USDC if you'd rather, and set the offer against the top of the local market.
• Equity: 0.60% of Saifuro, vesting.
• Health: 100% of medical, dental and vision premiums for US employees, 75% for dependents. Outside the US, the local plan through our employer of record, or $300 a month toward private cover.
• 25 days of paid vacation plus your public holidays, and we close between Christmas and New Year.
• 14 weeks of paid parental leave, for any parent.
• A laptop and $1,000 toward your workspace.
• $1,500 a year for learning, conferences included. That's on top of the events you'll attend for work: six to eight a year, planned a quarter ahead, on the company.

Hours are flexible. The one thing we ask is to be online from 14:00 to 17:00 UTC, when the team overlaps for calls and quick questions. That's a window, not the working day. The rest of your hours are yours to arrange. A mainnet incident doesn't keep office hours. If that window is the middle of your night, say so and we'll work something out.

You can combine this with a job at another company, as long as the two don't conflict.

How to apply

Email contact@saifuro.com with the subject "Machine Learning Engineer". Attach your resume and include links to your GitHub and LinkedIn profiles, as well as X or a personal site if you have them. Let us know your country of residence. We'll reply with the next steps.