A Y Combinator–backed startup says it's already processing more than $1 billion in credit operations through an AI platform that went live in mid-August, a figure that would be remarkable for a company with fewer than 10 employees. Levocred, based in San Francisco, deploys what it calls "AI agents" inside credit funds and banks to automate underwriting, portfolio monitoring, and compliance work across more than 30 facilities.
The timing puts Levocred in direct competition with an emerging class of AI-powered credit automation tools from much larger players. TD Bank rolled out mortgage automation on May 21. Fiserv launched its agentOS platform for banks the same month. Experian followed in June with its own Agent Operating System. Now a two-person founding team with private credit experience is betting they can move faster than the incumbents.
Mohit Gupta, who previously built AI models as a quantitative researcher at a credit hedge fund, co-founded Levocred with Saksham Gupta, who constructed trading and data infrastructure at Edge Focus as the credit fund's assets under management swelled from $150 million to over $1 billion. Mohit Gupta studied financial engineering at UC Berkeley. On LinkedIn, he described the platform as "the one I wished existed" during his hedge fund days.
The system handles tasks that typically consume hours of analyst time. According to the company's website, Levocred's agents draft investment committee memos from loan tapes and credit agreements, monitor borrower compliance in real time, and generate borrowing base reports formatted for lenders. The platform also tracks covenant compliance, provides treasury visibility, and assembles lender reporting packages.
"A loan tape and credit agreement go in; a cited IC memo draft comes out," the company explains on its homepage. The outputs cite source documents and are built for human review before final submission—a nod to the reality that no lender is likely ready to let AI make billion-dollar credit decisions without oversight.
A public demo posted at app.levocred.com shows the platform verifying a 4,182-row loan tape, running 12 eligibility tests, confirming an 85% advance rate, and drafting a CFO certificate. Whether that kind of automation translates into real efficiency gains depends on implementation, but at least one customer appears convinced.
Pier Asset Management, a Los Angeles private credit firm founded in 2017, both uses and invests in Levocred. The relationship offers validation, though it also raises questions about objectivity in the case study published on Levocred's website. Pier started with a single facility on the platform and now runs its entire book through the system, according to that case study.
Jonathan DiBenedetto, Pier's head of operations and chief compliance officer, said the platform reduced time spent managing the portfolio. "We hired our best analyst. It wasn't human," he wrote in a testimonial. Conor Neu, Pier's chief investment officer and co-founder, said the system handles scheduled reporting and routes alerts to Slack. "Levocred has become the operating layer of the firm," Neu wrote. "We spend our time on decisions now, not assembly."
That kind of language—"operating layer," "decisions not assembly"—reflects a broader shift in how private credit shops think about technology. Firms once built everything in-house or relied on legacy asset-based lending software from providers like ABLSoft, FinSoft, and Decipher Credit. Those platforms offer borrowing base automation and collateral monitoring but weren't designed with AI capabilities from the ground up.
Now a cluster of newer entrants is trying to stake out territory. Lama AI provides borrowing base automation for lenders. Fuse Finance markets what it calls "AI-native loan origination with multiple automation agents." TurnKey Lender offers AI-powered lending decisioning. A February whitepaper from Deloitte noted that AI agents in banking can handle document processing, fraud identification, and automated credit memo drafting—tasks that line up closely with what Levocred claims to automate.
Levocred argues that its governance features set it apart from general-purpose AI tools. "Agents read. They never write," the company wrote in its Pier case study, describing a design where agents analyze but do not modify underlying data. The platform isolates client data and cites sources for every number in its outputs, the company says—a design choice meant to address concerns about AI hallucinations or unexplainable results in high-stakes financial decisions.
The startup participated in Y Combinator's most recent batch and currently employs between two and 10 people, according to its LinkedIn profile. The company declined to disclose funding details beyond Y Combinator's backing and Pier's investment. For a company claiming to process over $1 billion in credit operations, that's a notably lean team—though perhaps exactly the kind of efficiency the platform itself promises to deliver.
