Teaching artificial intelligence to evaluate mortgage applications required more than software engineering.
When Pavan Agarwal and his team at Celligence LLC built AngelAi®, they trained the system using hundreds of thousands of real-world loan files combined with the full library of lending regulations that apply from the federal and agency level down to individual states and counties. Those rules change depending on the loan program, the property and the jurisdiction involved, so the team broke the underwriting process into smaller components and trained the system against each piece individually. A single loan application can require different documentation, eligibility standards and regulatory treatment depending on the borrower, property type and program, and Celligence developed individual methods around each part of that process before combining them into one unified system.
Agarwal’s own background shaped how that process was approached. Before founding Celligence, he worked inside Sun West Mortgage Company, Inc. in several roles, including as a wholesale representative, loan originator and software engineer, where he built ReverseSoft, a backend platform for reverse mortgage operations later adopted by lenders beyond Sun West. Those roles exposed him to the operational and regulatory detail the AI system would later need to reproduce.
AngelAi was also developed inside an operating mortgage company rather than as a standalone technology project. Before its commercial release in 2019, the platform was tested against live lending workflows within Sun West, allowing the engineering team to refine the system using practical lending operations alongside regulatory requirements.
The result is a platform built on what the company calls a deterministic architecture. The same input produces the same output every time, and every decision AngelAi makes can be traced back to the specific data and regulation that produced it. Sun West fully warrants every underwriting decision the system generates, a step Agarwal has said would not be possible without that traceability built in from the start.
The platform also incorporates Empathetic Technology, which operates alongside the underwriting engine during the customer experience rather than influencing lending decisions. According to the company, the system can recognize signs of confusion or frustration during the application process and connect borrowers with a human executive when additional assistance is needed.
Celligence’s engineering work has since produced more than 130 patents spanning artificial intelligence, blockchain and lending technologies, with an IP portfolio the company values at $119 billion. That portfolio grew out of the same development process used to train AngelAi on real loan files and lending regulations, giving the platform a technical foundation built around repeatable decisions, regulatory consistency and traceable results.
Today, AngelAi facilitates lending decisions across FHA, VA, USDA, conventional, and reverse mortgage products. The platform operates in more than 100 languages and has facilitated more than $40 billion in consumer finance across more than 210,000 transactions and serves over 408,800 registered users, a scale Agarwal has said reflects the years of internal testing that went into training the system before it ever reached a borrower.