Founding Engineer

Build the systems that verify, underwrite and collect the invoices behind $7 trillion of American receivables. Every small and big business in America has invoices sitting unpaid. $7 trillion of them. A business can take an unpaid invoice to a lender, get most of its value in cash now, and the lender collects later when it’s eventually paid. It’s how small businesses bridge the gap between doing the work and getting paid. In countries around the world, more than 10% of GDP runs through this kind of financing. In the US, less than 2%. It’s stuck because a lender can’t just hand over cash for all $7 trillion of invoices with no proof, they have to verify each invoice is genuine, the work was actually done, and judge whether it will be paid. Today that’s done by people, manually, on only the largest and easiest invoices. And fraud is rampant - BlackRock lost $430M to invoices that were completely fabricated. The software to automate this at scale doesn’t exist yet. That’s what Iridium is building. Live with US lenders today, backed by strategic institutional investors. You would work directly with Preesha, our CTO, on an engineering team where you’ll be given high responsibility. You own systems end to end, and you own the lenders who use them - onboarding, expansion, and the calls where you find out which of your assumptions survive a credit team. There is no layer between engineering and the customer here, and this is an opportunity to have full ownership over what you build The work is unforgiving in a specific way: a verification that is wrong is money gone. Anthony, our CEO, automated credit underwriting for banks and PE desks, and before that moved Morgan Stanley’s payments platform onto Kubernetes. Preesha, our CTO, built the recommendation models in Bloomberg’s terminal and has published twice on causality in large-scale datasets at tier-1 AI conferences. Knowledge ingestion engine: A huge portion of your work will be building the engine that makes verification possible at scale. Every signal - blurry receipts, Excel sheets, portal data retrieved by browser agents, payment histories, fraud indicators - has to be ingested, normalised, and synthesised into a single coherent picture of whether an invoice is real and whether it will be paid. This is the foundation everything else runs on. Portal agents: Building a general, self adapting browser agent, that can navigate across over 1500 portals to pull the approvals a verification depends on. Through 2FA, constantly adapting portal UI and at the scale of verifying 10s of thousands of invoices at once. Risk engine: A living view of the lending book that updates on every invoice funded and every payment received, instead of a credit decision frozen at origination. Greenfield. Voice collections: You will build the voice layer that collects on invoices the way a great human collector would. Not automated reminders. A voice agent that knows the history of the debtor relationship, understands the right moment to call, and has the judgment to adapt the conversation in real time. Evaluation: Extending our eval suite. Nothing ships on vibes, and this is why our customers trust us as the best-in-class. You have taken something from zero to production and owned it after it shipped. You have experience in any of these areas, and understand the challenges building the following would involve: You have worked at a startup and know how important to the company, a founding engineer is. You know what a cliff is. Strong computer science fundamentals. We index heavily on what you’ve built.