Frizzle Gives Teachers Back Their Weekends With AI-Powered Math Grading
Frizzle, a Y Combinator Summer 2025 company based in San Francisco, has built an AI product aimed squarely at one of the most stubborn time sinks in American education: grading. Teachers routinely spend more than 10 hours per week – roughly a quarter of all their working time – evaluating student work, especially handwritten math assignments that resist automation from older edtech tools. Frizzle’s AI grading software for teachers grades those assignments in minutes rather than hours, then turns the resulting data into rich analytics and differentiated assessment that teachers can use to personalize instruction student by student. The company was founded in 2025 by Abhay Gupta and Shyam Sai.
The founders describe Frizzle as the most valuable piece of software a classroom teacher can adopt, and the framing is earned by a simple observation: if grading goes from ten hours a week to a fraction of that, the recovered time is enormous on a personal level and transformative on a system level. Teachers can design better lessons, meet more students one-on-one, and work sustainable hours – a factor that has become increasingly urgent as burnout and attrition ripple across the profession. Beyond pure time savings, Frizzle’s ability to process every handwritten response and surface analytics on class-wide misconceptions means that instructional decisions can be informed by data that, under the old regime of manual grading, never made it to the teacher’s desk in any usable form.
Abhay Gupta, Frizzle’s co-founder and CEO, spent earlier parts of his career as a product manager at Coinbase, Tesla, and Meta, where he drove $50 million of incremental revenue at Coinbase and helped scale an online e-learning platform from zero to $5 million in ARR. He holds a bachelor’s degree in computer science and economics from Vanderbilt University. His co-founder Shyam Sai, Frizzle’s CTO, worked as a machine learning engineer at Microsoft, Meta, and Google Life Sciences before starting the company. That combination – a product leader with direct edtech scaling experience and a machine-learning engineer with a biomedical-AI background – maps neatly to the technical challenges of recognizing messy handwriting, reasoning about mathematical work, and shipping teacher-friendly classroom software.
Education technology has long struggled with AI because classroom realities are unforgiving: handwriting is inconsistent, student work takes many forms, and mistakes in an AI system can quickly undermine teacher trust. Frizzle’s bet is that modern multimodal models, paired with disciplined product design and domain-specific tuning, can finally clear that bar. Its near-term focus on handwritten math is strategic – math is one of the most universally assessed subjects, tests of mathematical understanding produce data that is genuinely rich, and the marginal cost of grading is high enough that schools are willing to adopt tools that reliably reduce it. If Frizzle nails that wedge, expansion into additional subjects and assessment types is the natural next move.
Operating out of San Francisco with a two-person team, Frizzle is part of the Y Combinator Summer 2025 batch and sits in the Artificial Intelligence, Education, SaaS, B2B, and Productivity categories. Its product is available at frizzle.com, and the company’s early focus is on teachers and schools who are willing to let software handle the mechanical parts of grading so that educators can concentrate on the irreplaceable human work of teaching. For a profession that has been asked to do more with less for years, AI grading software for teachers is one of the clearest trades in edtech: give Frizzle the grading, and get your evenings back.