The desks were next to each other in seventh-grade science. Years later, Aryo Patel and Tinah Hong would be sitting together again, this time trying to make software understand the paperwork of public companies. Between those two arrangements of furniture lay high school, MIT and a rather consequential discovery: a friend who tells you where your plan will break can be an excellent person to build it with.
When Patel invited Hong to work through a project in 2024, she devoted much of their first call to identifying the weaknesses in his approach. He remembers her being right. By the end of their first day working together, he had his answer about a co-founder. Chemistry is useful. Someone willing to inspect your assumptions is useful on a Tuesday afternoon, too.
Today Patel is co-founder and CEO of Andera, a San Francisco company building AI software for internal audit. Hong is its co-founder and CTO. In June 2026, the company raised a $37 million Series A led by Lightspeed. The number makes a tidy announcement. Their route there is more interesting: an abandoned idea, an education conducted through cold outreach, and a business built around work that resists a tidy answer.
The foil fencer who studied market crashes
Patel’s hometown is Wilmette, Illinois, and he attended New Trier High School. Before enterprise software supplied his competitive arena, he fenced foil. At New Trier he was a two-year captain, a Midwest Open champion and a state runner-up. He also founded a cybersecurity club. The interests were already fairly specific.
At MIT he studied mathematics and fenced for the Engineers. His 2021-22 season included a 15-1 overall record and an unbeaten 12-0 record in Northeast Fencing Conference competition. Those are results on a strip, with an opponent and a score. They do not explain his later business decisions, but they put a little flesh on a biography otherwise liable to disappear into software job titles.
He earned USFCA All-Academic and Scholar of Distinction recognition in three successive academic years, from 2020-21 through 2022-23. Asked about favorite student-athlete moments, he singled out the gatherings after tournaments. The record contains both competition and company.
His technical interests extended into finance. He undertook undergraduate research with Andrew Lo on indicators of market crashes, and held software engineering internships at Floating Point Group and Microsoft. He later worked on the trading desk at Jane Street. Finance offered him problems about evidence and uncertainty well before audit became his occupation.
Overall fencing record
12-0 in conference competition
Some of his earlier projects were aimed at making things easier to navigate, literally. At Hack STL in August 2019, his team won a “Most Practical” award for an interactive map helping New Trier students find their way between classrooms. Another high-school hackathon project explored signing into websites by selecting an image, using steganography and encryption. These were small, bounded problems with recognizable users: students trying to get somewhere, people trying to log in.
His public volunteer record also includes co-founding Family Focus Music Tutoring, a club providing music education to children in Evanston, and the Last Mile Connectivity Foundation, with a goal of bringing internet access to isolated villages in India. The activities put music, education and connectivity alongside the mathematics and code. They also widen the picture of what interested him before a company began to dominate the biography.
There is a pleasing time capsule on his personal website. It introduces a freshman looking to improve his computer science skills and invites visitors to consider him for projects and internships. Its friendly opening survives alongside a career that has moved considerably further along. An old personal site can be a more revealing artifact than a freshly polished biography: here is someone still asking for a chance to make something.
First, a rehearsal for difficult conversations
Andera was not Patel’s first attempt to turn AI into a practical working tool. In 2023 he joined Kevin Jiang as a co-founder of Simulation Labs, a platform for creating and playing AI simulations without writing code. Users could rehearse negotiation and conflict resolution, with a conversational model supplying the other side.
Patel described more than 4,000 users and scenarios ranging from negotiating a promotion to an extravagantly difficult customer. The proposition had an appealing economy: practice the awkward conversation before the real person arrives. A mistake in rehearsal is usually cheaper than one made across a desk.
The platform also found educational uses. Harvard Business School’s Michael Wheeler thanked Patel and Jiang for access to a negotiation exercise in which the user faced a simulated kidnapper. Separately, a professional mediator developed an ombuds scenario on Simulation Labs. It was an early encounter with a question that would return in another form: how do you make a general-purpose model useful inside a professional activity with its own expectations?
A good idea, crossed out
In April 2024, Patel secured backing from Pear before he had a co-founder or a product. The idea then was called RegCheck: software to help bank compliance officers follow regulatory changes. Interviews established a problem. They did less to establish a purchase. Reaching the people authorized to buy proved difficult, and the inconvenience did not seem urgent enough.
By June, he had decided to abandon it. Founders are encouraged to have conviction; the awkward part is deciding when conviction has become an expensive attachment. Patel changed his answer.
He moved toward SOX control testing and began contacting the people who understood it. The outreach reached about 4,000 cold emails a week and roughly 250 substantive discovery calls with CFOs and finance vice presidents. That is a considerable number of invitations to be ignored. It also gave him repeated chances to learn where the work hurt and who had reason to fix it.
The first idea gave way to a different, recurring job inside finance.
Seventh-grade trust, enterprise consequences
The partnership with Hong brought a long acquaintance into a new setting. They had attended the same high school and MIT. Hong’s experience included machine learning work at Stripe and software engineering at Microsoft. The friendship supplied history; the project supplied a test of how they worked.
Patel’s recollection of their early operating choices is unusually concrete. They built the product themselves, signed an initial contract worth more than $1 million and deliberately concentrated on one customer. In another episode, a promised feature required the pair to process files manually until six in the morning for two weeks. Automation can have a surprisingly manual childhood.
That customer concentration makes sense in the setting they chose. An auditor needs a result that can be checked, with evidence attached and decisions documented. More customers bring more revenue opportunities, but also more varieties of evidence, more exceptions and more promises. Their early choice was to stay close to one demanding implementation.
The evidence has to fit through the door
The engineering problem came wrapped in ordinary office files. Audit evidence could include sprawling Excel workbooks, embedded pictures, emails and contradictory PDFs. Reading the cell values alone would leave parts of the document behind. A screenshot pasted into a sheet might carry information that mattered to the test.
Andera built its own Excel engine, an effort that grew to roughly 43,000 lines of code. Its July 2026 engineering account describes the progression from extending existing libraries to replacing them. Among its reported results: workbooks that previously took two hours to load could load in less than a minute. The benchmark belongs to the company’s own experience, rather than a universal promise.
Patel describes the broader challenge as finding the small portion of a vast evidence collection that is relevant to the task.
“The hard problem is getting from billions of tokens to the 30,000 that matter.”
Aryo Patel
That observation gives the CEO’s work a useful specificity. A fluent answer is easy to admire. A conclusion connected to the right piece of evidence is what an auditor can use. For Patel, selling AI into this profession means making that connection dependable enough to enter someone else’s working day.
Who gets a seat beside the engineers?
Andera’s team combines auditors and engineers, intentionally keeping both disciplines close. Carina Averilla, its first hire, had worked at Deloitte and headed US accounting at Revolut. Jared Lauber brought decades of audit and risk experience and leads partnerships. Their backgrounds bring the buyer’s working habits into the room where the software is made.
The product follows that arrangement. Andera gathers evidence, tests controls and produces workpapers for review. Its GRC platform links risks, controls, narratives and issues, while giving reviewers the reasoning behind a result. Ambiguity becomes an item for a person to resolve. A useful audit tool has to accommodate the professional who remains responsible for the conclusion.
In August 2026, Andera announced an alliance integrating its technology into RSM’s next-generation internal audit platform. In September it introduced Andera GRC, responding to customers who wanted their wider audit program in the same environment as their testing. Those developments extend the job Patel first investigated through calls into a broader product and a relationship with an established accounting firm.
The aspiration is financial oversight that keeps pace with the business. Andera’s stated case for that work begins with trust: investors need confidence in company numbers, and verification should become easier to perform without becoming less rigorous. It is a substantial ambition attached to some decidedly unglamorous files.
Patel’s story has room for both. There are financing announcements and workbooks, a trading desk and a fencing strip, new customers and an old friend. The recurring choice is to get close enough to the work to discover what must change. Sometimes that means crossing out an idea. Sometimes it means moving the desks back together.