The San Francisco startup builds AI voice agents that take payments, chase late accounts and qualify loans - and grades every call against federal lending law. It has quietly run more than a billion dollars in payments through the line.
There is a specific kind of dread reserved for calling your mortgage servicer. The hold music. The transfer. The agent reading from a script, one eye on a compliance checklist, the other on the four other chats open on their screen. Kastle, a two-year-old startup out of San Francisco, looked at that experience and decided the fix was not a better script. It was a different kind of caller entirely - one that never sleeps, never multitasks, and knows the Fair Debt Collection Practices Act by heart.
Kastle builds AI voice and messaging agents purpose-built for consumer lending, starting with the least glamorous corner of it: mortgage servicing. Its agents answer inbound calls, take payments, walk borrowers through escrow questions, follow up on delinquent accounts and qualify new loan inquiries. Every conversation is transcribed and scored against the rulebook that governs lending in America. The company came out of Y Combinator's Summer 2024 batch, and in the time since it has done something most AI demos never do - it moved real money. More than a billion dollars of it.
Mortgage servicers - the companies that collect your payment every month long after the loan closes - operate enormous contact centers. When rates move or hardship spikes, call volume surges and staffing does not keep up. Borrowers wait on hold to do something as simple as make a payment. Agents work long hours juggling several conversations at once, and quality slips right when it matters most. The work is repetitive, emotionally heavy, and wrapped in regulation that punishes mistakes.
That combination - high volume, high stakes, high compliance load - is exactly the profile Kastle went hunting for. It is the sort of workflow big enough to matter and boring enough that few founders bother.
"We needed a partner with both proven performance and deep domain expertise at scale. Kastle stood out because they are specialized in the mortgage servicing space." Elizabeth Balce, EVP of Loan Servicing, Carrington Mortgage Services
Kastle's pitch to servicers is built on a handful of operational metrics. The headline is cost: it says its agents run roughly 70% cheaper per call than a staffed seat, while trimming average handle time and covering nearly every interaction with automated quality review instead of the sampled spot-checks most call centers rely on.
Rather than a single do-everything bot, Kastle ships a roster of agents mapped to the roles a servicing floor already staffs. Each one is narrow on purpose.
Handles payment reminders, takes payments, and answers routine questions on escrow, balances and statements, around the clock.
Engages delinquent borrowers and negotiates within compliance guardrails to bring past-due accounts current.
Scores calls and documents against the regulatory rulebook, replacing sampled reviews with near-full coverage.
Qualifies new inquiries, generates warm transfers and clears the manual follow-up loan officers dread.
A Loss Mitigation Agent extends the same approach into hardship and modification workflows, and an omnichannel layer keeps the conversation consistent across voice, SMS, email and webchat - handing off to a human the moment a call needs one.
The reason a general-purpose voice bot cannot simply wander into mortgage servicing is the law. Kastle's agents are pre-trained on the alphabet soup that governs lending - FDCPA, TILA, UDAAP, RESPA and TCPA - so the guardrails are built in rather than bolted on afterward. The company carries SOC 2 Type I and Type II certification and describes itself as GDPR-ready.
The other half of the moat is plumbing. Kastle integrated directly with ICE's MSP, the servicing system of record that a large share of the industry already runs on, plus consumer-direct CRMs and the major contact-center platforms. That means a servicer can go, in Kastle's framing, AI-native without ripping out the stack it depends on - the difference between a pilot and a production deployment.
"Carrington is setting the standard for what AI-native mortgage servicing should look like at scale." Rishi Choudhary, Co-Founder & CEO, Kastle
Kastle was founded in 2024 by Rishi Choudhary and Nitish Poddar, who met at the University of Illinois Urbana-Champaign. Choudhary, the CEO, became the founding product manager for Redfin's Mortgage Marketplace at 22 and grew it from $3M to $6M in revenue in nine months - close enough to the industry's pain to know where the bodies were buried. Poddar, the CTO, patented Verkada's motion-search system and was that company's youngest tech lead at 23, with a background in backend inference infrastructure.
The pairing shows up in how Kastle sells: deep enough in mortgage to speak the language of a servicing EVP, technical enough to plug into the systems those EVPs already run. The team is small - around a dozen people - and leans on a forward-deployed model, embedding engineers with customers to ship inside a regulated industry.
There is no shortage of companies building AI voice agents right now, and plenty of contact-center platforms are adding conversational AI. Kastle's wager is that horizontal tools stall at the door of a regulated industry, and that the incumbents - offshore BPOs and legacy call centers - are the status quo it is displacing. Its differentiation is specialization: agents built for mortgage compliance, wired into mortgage systems of record.
The customer base reflects that focus - US mortgage servicers, FDIC-insured banks and independent mortgage banks. The marquee name is Carrington Mortgage Services, which manages more than $200 billion in unpaid balances across over a million loans and, in mid-2026, moved its servicing operation onto Kastle's agents. Kastle describes the milestone as a template for what AI-native servicing looks like at scale, and calls its longer-term ambition an "operating system for consumer lending" - mortgage first, the rest of the lifecycle later.
The unglamorous parts of finance rarely make headlines, which is part of Kastle's advantage. A billion dollars in mortgage payments moved through an AI agent without a viral demo attached. Whether it becomes the operating system it is aiming for depends on the next set of servicers - but the phone queue, at least, has a new voice on it.