Andrea Pasinetti and the Lesson That Wouldn’t End

A school can have walls, desks and a place on a government map, and still be missing the person who makes it a school. Andrea Pasinetti encountered that problem in rural China. At a remote village in Yunnan, children crowded him with questions. Their curiosity was plentiful. Their chances of continuing their education were considerably less so.

The headmaster’s difficulty was keeping teachers. Distance and difficult conditions made the village a hard posting. Pasinetti had arrived as a student interested in public policy; here was a policy problem with children sitting inside it. He began looking for a way to help.

“They were really more eager for good teachers instead of more school buildings or multimedia classrooms.”

Andrea Pasinetti, 2013

He called home and Princeton to explain his plan. In 2008, he left university to build what became Teach For China. The thesis could wait. The children were rather less conveniently scheduled.

Three hundred schools, one unfinished degree

The journey had begun with language. Pasinetti attended Princeton in Beijing in the summers of 2005 and 2006. Studying Mandarin gave him a way into conversations that a visitor might otherwise miss. At Princeton, his interests included public and international affairs, East Asian studies and European cultural studies. China became a place to investigate educational opportunity at ground level.

During his research, he traveled through seven provinces and visited more than 300 schools, speaking with students, teachers, principals and officials. That is an unusually large detour from a university reading list. It also helps explain why the resulting organization concentrated on people: graduates from China and the United States, recruited and trained to teach in under-resourced schools.

Teach For China, initially called the China Education Initiative, took an idea familiar from Teach For America and brought it into another national context. The basic commitment was substantial: teaching in schools where attracting and retaining staff was difficult. Good intentions still had to survive the timetable.

By 2013, the organization had 210 teachers serving more than 30,000 students. Pasinetti was working in Beijing, fluent in Chinese, going from meeting to meeting in jeans and a T-shirt. At that point he said he had gone five years without a salary, sometimes paying work travel expenses himself. The expensive part of conviction can be remarkably literal.

Over the organization’s first five years, more than 450 fellows served in low-income schools and reached more than 70,000 students. Recognition followed, including China Newsweek’s 2011 distinction for his work in China and membership in the World Economic Forum’s Young Global Leaders community. The organization had become much larger than an undergraduate’s interrupted research project.

Back to class, with a different question

Pasinetti eventually returned to finish his Princeton degree in 2018. He went on to earn an MBA at Stanford and pursue graduate computer science studies. His route through education involved a lengthy period organizing it for others before returning to the student’s side of the desk.

There is a revealing little artifact from this return: an optional Stanford CS109 project submitted in June 2020. Pasinetti tried using a Naïve Bayes model to predict educational outcomes from school staffing ratios and capital spending in New York State. The question had changed countries, but the subject would have been familiar to anyone who had followed his earlier work.

He reduced the available records to about 780 complete entries, using 500 for training and 280 for testing. His write-up was candid about missing data, rough proxies and programming limitations. It reported roughly 54 percent predictive success, alongside qualifications about what had been left out. This was a class exercise, with the seams showing. There is something useful about keeping the seams visible.

In a later public tribute, he credited Stanford instructor Chris Gregg with helping him on his computer science journey. Gregg subsequently became part of Kira. For someone building teaching tools, having been a grateful student is a relevant qualification. So is remembering that understanding a subject and helping another person understand it are separate jobs.

The intern who became a co-founder

Kira’s early team gathered in Palo Alto in August 2021. In Pasinetti’s account, Andrew Ng’s guidance gave the young venture an advantage from the outset. Jagriti Agrawal, then between years of her Stanford MBA, was still an intern.

She had already worked at NASA’s Jet Propulsion Laboratory on automation software for the Mars 2020 rover mission. Her contribution to Kira soon exceeded the label on her position. In September 2022, Pasinetti publicly announced that she would formally share the co-founder title, recognizing work she was already doing.

It is a small episode with a clear view of how he described his colleagues: specific about their contributions, generous about their ability, prepared to change the organizational chart to match reality. Startups have a great fondness for titles. Here, at least, the title had some catching up to do.

Ng became chairman. His AI Fund and New Enterprise Associates backed Kira’s $6 million seed round in 2022. A May 2023 securities filing recorded a further approximately $15 million offering, signed by Pasinetti as CEO. There was now venture capital behind an undertaking that had begun with questions about teachers and access.

A browser extension with a larger assignment

The original Kira was modest in form. Pasinetti has described it as a browser extension intended to help people teaching technical subjects without a deep background in those subjects. It began before today’s commercially available large language models, using an earlier natural-language-processing approach.

That origin matters. A teacher newly assigned computer science has two subjects to learn: computer science, and how to teach computer science. Any tool entering that room has to earn its place with someone already doing a demanding job.

When NEA first discussed the venture with Pasinetti in fall 2021, the idea was a STEM platform with current material and room for teachers to adapt it. The early product combined instructional content, coding exercises in the browser, progress information and professional development. Keeping the coding environment inside the browser removed a particularly unglamorous obstacle: setting up software before the learning could begin.

Pasinetti’s technical work also extended beyond classrooms. In 2022, he was identified as co-founder and head of engineering at VIAVIA, a fashion commerce business led by Sixuan Li. He described an AI product-development tool connecting designers and brands with manufacturers. AI Fund now lists him as head of engineering and co-founder at BNTO, alongside Li and other founders. Education remains his public leadership role at Kira; engineering is also part of his work elsewhere.

The distance between a lesson and its learner

In 2023, Pasinetti and Agrawal traveled to Togo to sign an agreement with digital economy minister Cina Lawson. The aim was to make foundational computer science available to university students regardless of their major. By June 2025, Kira described a training effort involving 44 lead tutors, a planned network of 900 course tutors and an initial target of 15,000 students at the universities of Lomé and Kara.

The numbers describe a rollout plan, rather than a completed outcome. The structure is the interesting part: educators training educators, with software underneath. Even in a business centered on AI, distributing knowledge still involves people who know the local classroom.

In an August 2025 conversation about personalized learning, Pasinetti argued for instructional materials tailored to individual students. That moves the ambition beyond giving everyone the same digital textbook. A student’s previous answers should help determine the next useful explanation. It is an appealing proposition, and a demanding one to implement responsibly in an actual class.

What should happen after the wrong answer?

Kira launched its 2.0 platform in New York on March 3, 2026. Its wider brief brought course creation, tutoring, assessment and learning information into a connected system. Teachers would be able to inspect a student’s progress and generate follow-up work within the same environment. Grades remained subject to teacher review before release.

The distinction between an answer and an understanding has become more explicit in Pasinetti’s own thinking. In August 2026, he described a small internal study involving high school students preparing for the SAT or ACT. The team compared Kira’s estimates of students’ knowledge with independent diagnostics. He pointed to encouraging alignment, while explicitly resisting the idea that this proved the learner model correct.

“The harder question is what it understands before it produces anything.”

Andrea Pasinetti, 2026

That caution belongs in the story alongside the ambition. A system can produce a convincing lesson quickly. Establishing that it has correctly identified what a particular student needs is another obligation, and a small internal study leaves plenty to test.

September’s product updates added resources based on students’ specific errors, student goals and assessment diagnostics. By October 20, Kira was describing how teachers could turn the resulting learning information into targeted practice, then use the next round of student work to reconsider the next step. The teacher remains responsible for decisions about support.

Pasinetti’s career has moved from recruiting graduates for distant schools to building software that attempts to notice differences between individual learners. The methods have changed considerably. The attention to what reaches a child has persisted. Leaving Princeton was one way of following that question. Returning to class was another. Kira is the current attempt, and the lesson is still in progress.