Pavan Agarwal Is Betting on a Future Where Financial Software Requires Less Navigation

Pavan Agarwal wants to challenge one of the oldest assumptions in financial software, that users should have to learn the system before the system can understand them.

The idea sits behind AngelAi's Zero UI concept, introduced in January 2026. The company describes it as a way for users to express what they want in natural language while the system interprets the request, determines the appropriate action, and handles the underlying workflow.

Agarwal laid out the thinking behind the concept in his article on National Mortgage Professional about what he called the mortgage industry's interface problem. He pointed to the familiar sequence of opening screens, finding forms, entering information, correcting errors, and moving through a prescribed workflow. He argued that mortgage technology should move toward intent, transactional AI, adaptive interfaces, and human assurance.

AngelAi's Roshni release, announced in February 2026, puts that idea into a working product. The interface is designed to adapt the way information is presented according to the task, device, and context. Users can communicate their goals naturally, while AngelAi® interprets their intentions and works through the relevant business rules and regulatory requirements.

For financial software, understanding a request is only the beginning. A system also has to execute the resulting action correctly and maintain a record of what happened.

Sena, released by Celligence LLC, in August, pushes AngelAi further in that direction by adding Angel Agents built to execute delegated tasks rather than simply interpret requests. The release includes agents for areas such as email, calendar management, research, and document work, extending AngelAi's conversational model into broader day-to-day execution.

AngelAi's Zero UI whitepaper describes a Transactional Language ModelTM that converts natural language requests into structured, schema-bound actions. It also describes a confidence-gated system that can seamlessly have a human review uncertain cases. The company says the design is intended for transactions that require accuracy, compliance, and auditability.

Mortgage lending involves borrower information, financial calculations, and regulated processes, all of which have to be handled within the requirements of the loan being processed.

Agarwal has developed AngelAi within Sun West Mortgage Company, Inc. (NMLS 3277), where he serves as CEO. The platform was developed and tested inside the lender before its commercial release in 2019, giving the technology a live mortgage environment in which to be developed.

Roshni also changes the role of the interface itself. Instead of presenting every function through a fixed set of screens, AngelAi describes an experience that can adjust what the user sees according to the work being performed.

A mortgage professional could therefore communicate the task first and let the software determine how that task should be handled, rather than learning the application's preferred sequence beforehand.

Agarwal described the purpose of Roshni when it was announced in February.

“With Roshni, users can directly experience the intelligence that has always powered AngelAi,” Agarwal said. He added that the release was intended to improve clarity, responsiveness, and efficiency while maintaining the deterministic execution and compliance standards required for high-stakes financial transactions.

The underlying architecture is designed to keep the conversational layer connected to controlled financial actions. AngelAi describes its transactional model as deterministic, with defined rules and structured actions governing what the system can execute. The company says this allows a natural language interaction to lead to an auditable transaction rather than simply producing a plausible answer.

A user can ask what a mortgage condition means and receive an explanation. A transactional system has to determine what action is permitted, apply the relevant rules and execute it correctly when the user asks for something to be done.

AngelAi's Zero UI model puts more of that work inside the software.

The user communicates the intended outcome. The system interprets the request, applies its permitted logic and presents the result through an interface suited to the task. When the system is uncertain, AngelAi's architecture provides for human review.

The concept also connects with other work from Celligence. In December 2025, the company received a U.S. patent covering a conversational system for transferring digital assets. The patent describes natural language instructions being used to identify a recipient and amount, construct a transaction, and apply verification before completion.

For Agarwal, the broader question is how much of financial software's traditional navigation can be handled by the system itself.

Zero UI is his answer in development.

Whether the model spreads beyond AngelAi remains to be seen. What is already documented is the direction of the technology, with AngelAi moving toward financial software where users can describe the outcome they want while the system manages more of the navigation, validation, and transaction work underneath it.