A $5 Million Round for Skill in Action
DeweyLearn announced on 16 July 2026 that it closed an oversubscribed $5 million Series A led by SJF Ventures, with participation from Catalysis Capital, Morningside, and Owl Ventures. The company builds a multimodal AI platform that assesses how well someone performs a real task by analyzing audio, video, and learning data together, then delivering feedback at a level of detail that used to require a scarce human expert. The plan for the new capital is to extend that assessment across clinical and healthcare education, higher education, workforce training, and K-12 environments.
The premise addresses a long-standing constraint in learning. Judging whether a person can actually do something, as opposed to answering questions about it, has always depended on an experienced observer watching the work and giving feedback. That expertise is expensive and limited, so hands-on assessment rarely scales. DeweyLearn's wager is that multimodal models can now watch the performance itself, recognize the difference between competent and expert execution, and return specific coaching. If that holds, skill verification stops being a bottleneck gated by the availability of senior evaluators.
What the Model Actually Watches
The system has real production evidence behind the pitch. At the Auguste Escoffier School of Culinary Arts, DeweyLearn graded more than 20,000 student submissions, offering immediate feedback on knife technique and other practical skills and saving instructors hundreds of grading hours. Chief executive and co-founder Luyen Chou used that example to describe the ambition. "Human expertise has been a limited resource for our entire history," he said. "With multimodal AI, DeweyLearn can give an aspiring chef real-time feedback on her knife technique from the world's greatest chefs."
The applications extend well past the kitchen into regulated, high-stakes training. The platform assesses clinical therapist effectiveness, gives real-time feedback to nursing students working in simulated hospital environments, and supports continuing education for therapists through the NeuroAffective Relational Model. Chief technology officer and co-founder Dirk Liebich framed the underlying approach in scientific terms. "Much like Google Earth, we're building a world model of learning that empirically understands learning at an action level and can apply the insights at scale," he said. The unit of analysis is the observable action, not the written answer.
Why Investors and Judges Took Notice
The raise came with external validation that is unusual for an early-stage company. DeweyLearn won the 2026 ASU and GSV Cup, an award that recognized it as the top education technology startup among more than 3,000 companies worldwide. That kind of recognition, arriving alongside an oversubscribed round, signals that the approach resonates with both practitioners and investors who see a lot of edtech pitches. Arrun Kapoor, managing director at SJF Ventures, tied the investment to outcomes. "SJF is focused on enabling better learning outcomes and career opportunities for all," he said.
Kapoor also pointed to execution risk as the reason the firm was convinced. He said SJF was impressed by the transformative potential of DeweyLearn's approach to applying multimodal AI for education, and confident that the founders had the industry experience and AI expertise to deliver on it. That emphasis is telling in a market crowded with AI claims. The scarce ingredient is not the model but the domain knowledge and validated data needed to make its judgments trustworthy in fields where a wrong assessment carries professional and, in healthcare, human consequences.
The Assessment Bottleneck Enterprises Know Well
The problem DeweyLearn targets is one corporate learning teams recognize immediately. Compliance modules and multiple-choice quizzes are easy to scale because a machine can grade them, so that is what most training programs measure. Actual competence at a physical or interpersonal task, whether a technician performs a safety procedure correctly or a support agent handles a difficult call well, usually requires a supervisor to observe and judge. That human bottleneck is why hands-on skill verification is rare, inconsistent, and often skipped in favor of a certificate that proves attendance rather than ability.
Video-based assessment scored by multimodal AI offers a route around that constraint. An organization could ask employees to record themselves performing a task, then use a model trained on expert judgment to evaluate the performance and return specific feedback at a scale no team of human assessors could match. DeweyLearn's early deployments in nursing simulation and skilled trades adjacent work suggest the pattern generalizes beyond any single field. The prize for enterprise L&D is assessment that measures what someone can do, delivered often enough to actually improve performance.
The Trust and Governance Bar Is High
Automated skill assessment invites hard questions that buyers should ask before deploying it. An AI grading a nurse or a technician has to be accurate, fair across demographic groups, and explainable enough that a person can understand and contest a score. The stakes rise when the assessment gates a certification, a promotion, or a role. DeweyLearn's partnerships hint at how it addresses this. Katy Genseke, head of clinical product at Riverside Insights, said the assessment provider is working with DeweyLearn to explore new approaches that improve efficiency and effectiveness of its offerings.
For technology and learning leaders, the diligence checklist is concrete. Confirm how the model was validated against human expert judgment, whether it has been tested for bias, and how employees can appeal a result. Recording people performing tasks also raises privacy and consent obligations that need clear policy before any rollout. The upside of scalable, expert-level assessment is real, and it will only be realized by organizations that treat the fairness and privacy questions as prerequisites rather than afterthoughts once the tool is already in production.
What It Means for the Roadmap
The practical signal for L&D leaders is that skill verification is becoming a machine-scalable function, and the gap between measuring knowledge and measuring ability is starting to close. Teams planning training investments should watch this category and identify where hands-on competence matters most, in safety, customer interaction, or technical procedures, because those are the areas where video-based AI assessment could soon replace assessments that today either do not happen or happen too rarely to change behavior. Early pilots in controlled, high-value use cases will teach more than waiting for the market to mature.
The wider implication is a shift in what a credential means. When AI can watch the work and judge it, certifications can be grounded in demonstrated performance rather than seat time or a passing quiz, which raises the credibility of internal skills data across the enterprise. DeweyLearn is one funded example of a broader move toward evidence-based skills measurement. For CIOs and CHROs building workforce plans on skills, an assessment layer that proves capability, rather than assuming it, is exactly the missing piece that makes the rest of the data worth trusting.
