# DeweyLearn raises $5 million to grade real-world skills with multimodal AI > Source: > Published: 2026-07-18 12:21:25+00:00 [DeweyLearn](https://deweylearn.com/?ref=runtimewire) announced in a July 16th [PR Newswire release](https://www.prnewswire.com/news-releases/deweylearn-secures-5-million-in-series-a-funding-to-scale-ai-that-assesses-real-world-skill-at-expert-level-302827161.html?ref=runtimewire) that it raised $5 million in Series A funding, giving co-founders Luyen Chou and Dirk Liebich more capital for a bet that looks different from the last two years of chatbot-heavy edtech: AI that watches people perform real work, then grades the work against an institution's own curriculum and rubrics. The round was led by [SJF Ventures](https://sjfventures.com/sjf/?ref=runtimewire), with Catalysis Capital, Morningside, Owl Ventures and others participating. DeweyLearn described the financing as oversubscribed. It did not disclose valuation, revenue, pricing, ownership terms or the exact close date of the round. Chou is the reason this deal is more than another small AI education financing. He has spent his career inside the machinery DeweyLearn is trying to change. DeweyLearn says he previously served as chief learning officer at 2U, chief product officer at Pearson and founded The School at Columbia University. A 2018 Trilogy Education announcement adds more of the career arc: Chou joined Pearson through its acquisition of Schoolnet, held senior product roles there, and had earlier worked at The School at Columbia University and Learn Technologies Interactive before becoming Trilogy's chief product officer. The founder story runs deeper than a resume line. DeweyLearn is named after John Dewey, the philosopher and education reformer who, according to the company, encouraged Chou's grandmother to study teaching at the University of Chicago during Dewey's China trip from 1919 to 1921. [Teachers College, Columbia University](https://www.tc.columbia.edu/articles/2026/july/building-human-centered-ai-tools-for-a-new-era-of-education/?ref=runtimewire), where Chou earned a master's degree, framed DeweyLearn in a July 2026 profile around a human-centered AI thesis. In that piece, Chou emphasized keeping humans in the loop, a useful check on how DeweyLearn is positioning itself: as faculty augmentation rather than faculty replacement. ### The product is aimed at performance, not recall DeweyLearn's software combines audio, video and learning data with customer-specific curriculum, domain knowledge and assessment rubrics. The company says the system can observe performance in physical or online settings and produce expert-level feedback for fields where a human assessor would normally need to see, hear and interpret what a learner is doing. That makes DeweyLearn less like a study companion and more like assessment infrastructure. On its homepage, DeweyLearn says it is trained on an organization's curriculum and domain expertise, then watches learning in action to deliver insights at a scale human experts could not match. The company lists use cases across career training, culinary schools, counseling and classrooms, and says the more learning the system analyzes, the more it can surface patterns back to the organization. The sharpest proof point in the release comes from Auguste Escoffier School of Culinary Arts. DeweyLearn says it has graded more than 20,000 student homework submissions there, acting as a first review before instructor approval or adjustment. The company's [Escoffier case study](https://deweylearn.com/escoffier?ref=runtimewire) describes a workflow where the system analyzes live online classes, reviews student demonstrations through photos of cooking at home, and helps instructors grade complex multimodal assignments. That is a narrower and more credible claim than the release's broad assertion that DeweyLearn can handle any use case requiring expert observation. Culinary homework gives the system a domain with visual evidence, clear rubrics and repeated assignment patterns. The same logic may extend to nursing simulations, therapist training or classroom observation, but each new domain will test how quickly DeweyLearn can encode local expertise without turning every deployment into a services project. ### Chou's institutional route is the commercial bet DeweyLearn is selling into institutions with existing assessment burdens. The July 16th release names Auguste Escoffier School of Culinary Arts, Riverside Insights and the NeuroAffective Relational Model, or NARM, as users or partners. DeweyLearn's homepage also names Meteor Education, and the company's [Meteor Education case study](https://deweylearn.com/meteor-education?ref=runtimewire) describes classroom observation and coaching workflows that measure student engagement, collaboration and prosocial behavior. Teachers College reported that more than 2,000 culinary students were receiving targeted feedback through DeweyLearn and that the product was being used across corporate competency measurement, therapist training and behavioral assessment support. Those are company-facing and institution-facing deployments, not a consumer learning app chasing low-cost distribution. That route fits Chou's background. Pearson, 2U, Schoolnet, Columbia and Trilogy were all institutional education or workforce training businesses in different forms. DeweyLearn's Series A is therefore a founder-market-fit round: investors are backing a founder who knows how schools, training providers and education companies buy software, and a CTO, Liebich, whom the company describes as an applied AI, data analytics and predictive analytics expert. Liebich's framing is the more technical one. In the release, he said DeweyLearn is building a model of human learning through cycles of observing, intervening and measuring across domains. He said the system builds a customer-specific knowledge graph, then contributes to a larger "meta knowledge graph" of how people learn. The claim is ambitious; the commercial version is more concrete. DeweyLearn needs each deployment to get smarter locally while still producing reusable patterns across customers. ### The financing leaves the hard numbers outside the frame The $5 million Series A is the only new financing number DeweyLearn disclosed on July 16th. A separate 2025 [Form D summary posted by StreetInsider](https://www.streetinsider.com/SEC%2BFilings/Form%2BD%2BDeweyLearn%2C%2BInc./24428647.html?ref=runtimewire) showed DeweyLearn had sold $3,799,999 of a $3,999,999 equity offering to five investors. The Series A release does not say whether that prior exempt offering overlaps with today's financing, and it does not name the earlier investors. SJF Ventures' participation gives DeweyLearn a lead investor with an education and workforce thesis. The firm describes itself as an impact venture capital fund founded in 1999. Arrun Kapoor, a managing director at SJF Ventures, said in the release that the firm was backing the founders' combination of industry experience and AI expertise. For DeweyLearn, the money buys time to prove that its product can cross domains without becoming too bespoke. The named customer set is deliberately broad: culinary education, specialty assessments, clinical training, K-12 classroom environments and workforce learning. Breadth helps a young platform tell a bigger story. It also creates a product-management burden, because knife technique, therapist intervention quality, classroom talk and corporate language practice do not share the same evidence, workflow or buyer. The education AI market has been flooded with tools that generate content, tutor students or summarize classroom activity. DeweyLearn is making a different claim: institutions need to measure whether learners can actually perform. If Chou and Liebich can make that assessment layer repeatable, DeweyLearn could become embedded in the workflows that decide progression, certification and coaching. If the model requires heavy customization for every domain, the Series A will fund a slower services-heavy business. The honest read is that DeweyLearn has early deployment evidence, a founder with unusually relevant institutional education experience, and a timely wedge into performance-based assessment. The unanswered questions are the business ones: how many customers are paying, how much each deployment costs to launch, whether the software improves with scale across domains, and whether institutions will trust AI-generated assessments when those assessments affect grades, credentials or professional advancement.