Software Engineering Placement Job New York New York USA,Finance & Banking

Location: New York

Software Engineering Placement
· New York

Iridium is building the infrastructure that scales invoice finance. So that every business can access the money they’re already owed.

Backed by strategic institutional investors.

Context

Every business that sells to another business has invoices. Every time goods are delivered or a service is rendered, an invoice exists. In aggregate, US businesses are sitting on $7 trillion of this.

Contracted. Documented. Due.

It is, in theory, the most abundant form of collateral in existence. And yet less than 1% of it is ever used to secure financing.

Lending against invoices isn’t new. Lenders have known for decades that a confirmed invoice from a creditworthy buyer is about as safe a bet as commercial lending gets - you’re underwriting a transaction that already happened. But doing it safely is brutally, structurally manual, and that burden doesn’t stay flat as you scale. It compounds. To lend against an invoice, a lender has to do three things every single time:

Is this borrower legitimate, is the debtor actually going to pay, and what does the risk look like across the whole lending book when you add this invoice in?

Verify

Is this invoice real, was the work actually done, and does the backup documentation across email threads, supplier portals, and accounting systems hold up?

Collect

When does the money actually arrive, and who is chasing the invoices that aren't paying on time?

Lenders process thousands of invoices every week. Each one needs all three and there is no shortcut that doesn’t introduce risk. The temptation to sample, verify 20% and fund everything, is how the industry has lost billions.

Black Rock's HPS

$430M

lost to fabricated invoices that cleared multiple audit checkpoints.

$715M

lost to invoices that turned out to be misrepresented.

Fraudulent invoices look exactly like real ones, until they don’t.

Why nothing has fixed it

You’d think LLMs would have solved this by now. They haven’t.

Even if the models were good enough, the harness doesn’t exist.

No system that maintains a living view of a lending book and updates it as every new invoice changes the risk profile. No engine that gets smarter with every invoice funded and every payment collected. No voice layer that can collect on invoices the way a human would, with the right context, the right tone, and the right moment. Automated reminders recover a fraction of what a well-timed, informed conversation does.

What you will work on

One hard problem at the core of the company. Not a rotation.

Invoices don’t get paid the way the terms say they will. Every debtor has a rhythm - who pays early, who pays at 74 days regardless of what the contract says, whose behaviour shifts before they stop paying entirely. You will build the systems that learn that rhythm from real payment histories and turn it into something the rest of the platform can act on.

Collection Policy Generation

Once you can predict how an invoice will actually be paid, the question becomes what to do about it. You will work on generating the collections policy itself - when to reach out, through which channel, and how hard - and on feeding the same signal back into how we underwrite the next invoice. This is a live problem for our lenders, not a research exercise.

Across The Stack

Next.js and React on the front, Python and FastAPI on PostgreSQL and AWS behind it, with heavy use of agentic AI, evaluation pipelines, search and retrieval, and time-series data. You will not be handed one narrow slice. You will follow the problem wherever it goes.

Forward Deployed

You will get real exposure to the lenders we serve. Sitting in on how they work, seeing which of your assumptions survive contact with an actual credit team, and bringing that…