YC Root Access Sep 18, 2026
With Rishi Choudhary, co-founder and CEO of Kastle · Nitish Poddar, co-founder and CTO of Kastle
About 10 million Americans make their mortgage payment over the phone every month, and almost all of those calls are answered by a person sitting in a US contact center. That fact, picked up in a conference hallway in San Francisco, is the business Kastle now runs.
The founders had two other startups inside the same Y Combinator batch. They killed the second one, which was making money, a month before Demo Day.
"we didn't feel that this was existential for our customers"
Rishi Choudhary was the founding product manager for Redfin's Mortgage Marketplace before starting Kastle with Nitish Poddar, who he met studying engineering at the University of Illinois. The company has now crossed $2 billion in payments processed by AI agents and has just raised a $24 million Series A led by Insight Partners.
The full interview is covered here so you can skip it.
Here are the 10 numbers that matter.
Key Takeaways
Kastle works with 10 of the 25 largest mortgage servicers, automating customer service and collections
Volume has gone from about $10M every two months to that amount in a single day, on the founders' account
The product is positioned as an AI employee rather than a new system of record, because replacing a bank's core can take five to ten years
The idea they killed was earning $4,000 a month when they shut it down, a month before Demo Day
The insight that became the company: 10 million monthly mortgage payments made by phone, answered onshore
Their first big break was a pitch competition where the live demo failed twice before working
They moved the whole company into Newrez's office and stayed until the deployment went live
The same agents now run on auto loans, credit cards, personal loans and HELOCs, and on origination
The hard engineering problem is making non-deterministic models behave deterministically inside a regulated process
The constraint they say they remove is operational bandwidth, not demand for credit
1. What Kastle Does
Kastle sells what the founders call an AI employee: software that runs the operational work inside a bank rather than a new system for the bank to migrate onto. It started in one place.
The starting point was mortgage servicing
We started two years ago, starting with mortgage servicing operations, and today we work with 10 of the top 25 servicers
Kastle's founders
The work automated is customer service and collections across a servicer's business.
2. $2B Processed
Asked where the volume stands, the number came with a comparison that makes the rate of change legible.
The cumulative figure
So we recently crossed $2 billion in transactions processed using AI agents.
Kastle's founders
And what it looked like a year ago
it's crazy to think that same time last year we were doing about $10 million every two months, and now we do that every day.
Kastle's founders
3. Why An AI Employee
The founders' diagnosis of banking is that growth has a headcount problem attached to it.
Growing the business means growing the back office
I feel like most of the problems that banks face today is that, in order to grow revenue, grow loans, grow deposits, they also need to grow the back office.
Kastle's founders
And the second constraint is behavioral
And one of the interesting things that we saw is that banks are also really bad at change management.
Kastle's founders
That left them a choice, and the reason they took the one they did is the timeline on the alternative.
Replacing a core system is a decade-long project
It's very hard for such large, regulated industries to completely replace their core systems.
Kastle's founders
So they sell the value without the migration
help banks adopt AI in a reliable way and deliver value quickly, without them having to go through this entire process of changing their core systems, which might take five to ten years
Kastle's founders
The model came from their own experience of the tooling.
What they saw in their own workflow
We were seeing a lot of rapid improvements happen in — for example, coding, where we use tools like Devin — and they've helped us supercharge our workflows, and we found that banks really needed something similar, in order to supercharge their own operating workflows, but there was no good solution out there that was helping them do that.
Kastle's founders
The market as they found it
It was either you changed the existing system of record, or you had this one-point solution.
Kastle's founders
4. Two Ideas, Both Killed
The company went through Y Combinator in the Summer 2024 batch with something else entirely.
The application idea
I think it was real estate assistance, to help real estate agents close more loans, close more deals — AI for home buying, essentially.
Kastle's founders
It lasted a week
I think within week one we proved out that we had a pilot. We lost that, and then we realized this is not going anywhere
Kastle's founders
The second attempt was an AI sales development representative aimed at home buying, finding leads for brokers. They worked on it through June, July and into August. On the host's account it reached about $4,000 in monthly recurring revenue, and they shut it down a month before Demo Day, which put them back at zero for the second time.
5. Why They Went To Zero
Choudhary described the decision as a conversation in a car on the way to an airport. The test they applied was not whether the product worked.
The question that killed it
we didn't feel that this was existential for our customers
Rishi Choudhary
What that made them
It felt like we were a nice-to-have for them.
Rishi Choudhary
And why they started again anyway
And Nitish and I, we really care about building something for the long term, and we wanted to go back to the drawing board, no matter how painful that was, and restart over, to build something that our customers couldn't live without.
Rishi Choudhary
6. Thirty Days Of Conferences
With a month left, the search was physical rather than analytical: flights around the country, every conference they could find, and cold conversations in hallways.
What the search actually involved
It was a lot of awkward conversations, just standing outside conferences as people walked out and trying to cold-meet them and talk about their problems.
Kastle's founders
They said they got better at it, and less awkward, which is the part of the method worth copying.
7. 10M Phone Payments
The fact that became the company came out of one of those conversations.
The number nobody outside the industry knows
it was actually at one of the conferences here in San Francisco where we learned about 10 million Americans making their mortgage payment every single month over the phone.
Kastle's founders
The host said the figure shocked him. What matters commercially is who answers those calls.
Every one of them reaches a person
And all of these calls were actually being handled by a human agent in the contact center
Kastle's founders
The reason is regulatory. Offshoring the work is treated as too risky for a mortgage lender, so the cost base stays onshore.
Where the labor sits
So all of these calls were being taken in the United States, in places like Phoenix, Dallas
Kastle's founders
What the lenders said they wanted
and all the mortgage lenders could talk about was that if someone could solve this problem for us, that would be a game-changing experience
Kastle's founders
And why the window was open
and voice was relatively new at that time, and no one was really doing it to process payments for such a regulated industry as mortgage
Kastle's founders
8. The Demo That Failed
Two weeks from Demo Day, they entered a pitch competition in San Diego. Poddar built the demo from a corner of the auditorium while Choudhary worked the floor. The demo was a phone call that had to take a payment on stage.
It did not work, twice
First time he called in, it didn't work, on stage, in front of hundreds of people. So we blamed it on his phone — we said, okay, it'll work with my phone instead, so I gave him my phone. Second time we called in, it didn't work, and then I said, oh, I forgot to turn it on.
Nitish Poddar
The third attempt landed
And then the third time we called in, it actually ended up working.
Nitish Poddar
And the result was the opposite of what they expected
but it turns out we actually won that competition, and that created a ton of inbound for us and helped us land our first customer, in spite of a demo malfunctioning.
Nitish Poddar
The first customers closed a few days before Demo Day.
9. Moving Into Newrez
Early customers were prototypes. The one that made the company was Newrez, which the founders described as the third largest servicer in the country, and the method of landing it was not remote.
What it took to go live
we moved the entire company to Phoenix and moved into their office, and didn't come back until they were live.
Kastle's founders
10. Where It Goes Next
Asked where the company goes over five to ten years, the founders framed mortgage servicing as the proof rather than the market.
What the first deployment established
What we've been able to do is prove that applied AI autonomous agents can work in one of the most regulated operational environments in financial services, which is mortgage servicing
Kastle's founders
The technology built underneath it, which they described as making agents effective in real-time settings, is now being pointed at the rest of consumer lending. Large banks are using it to service auto loans, credit cards, personal loans and home equity lines of credit, and to originate loans as well.
The intended shape is a co-worker: agents taking the high-volume, labor-intensive work and leaving the cases that need judgment to people. Their claim about the market is that the demand side is not the problem.
The binding constraint is operational, not commercial
The only thing that stops them from doing that is them not having the operational bandwidth or resources to provide personalized experiences for each of their customers
Kastle's founders
Bonus Insights
The hard part is making a probabilistic system behave
The host named the engineering problem directly: Kastle lets customers prompt the agents themselves, the agents run on large language models that are non-deterministic by construction, and the industry they operate in requires the same input to produce the same compliant output every time. Building a harness that makes a customer's own changes deterministic and auditable, the host said, is difficult work, and the founders did not dispute the framing.
With the Series A closed and the largest banks in the country now in the pipeline, the founders said the money goes into three places.
Three teams get the Series A money
So we're growing our applied AI team, growing our product teams, as we get into multiple different products, and then also scaling our infrastructure teams.
Kastle's founders
The founders' bottom line is that the way into a regulated industry is to sell the output rather than a migration: leave the bank's core systems alone, take the highest-volume manual work off its contact center, and prove it in the most tightly regulated corner first.
Products, Companies & Tools Mentioned
Kastle (The AI employee for bank operations: 10 of the top 25 mortgage servicers as customers, $2 billion processed, and a $24 million Series A)
Newrez (Described by the founders as the third largest servicer in the country, and the customer they moved the company to Phoenix to deploy)
Insight Partners (Led the $24 million Series A)
Devin (The coding agent they use internally, and the model for what they think a bank's operations team needs)
Y Combinator (The Summer 2024 batch they entered with a different idea, and where they abandoned a revenue-generating product a month before Demo Day)
Watch the full episode:
More Rishi Choudhary interviews and podcast appearances
If this was worth your time, send it to someone who follows the name.
Get the latest market chatter as it happens: