DeweyLearn’s $5 Million Raise Meets a Hiring Trust Crisis

DeweyLearn raised $5 million on July 16 to scale an AI that grades how someone performs a real task, from a knife cut to a clinical exam. The Series A round was oversubscribed and led by SJF Ventures.

It lands at an odd moment for hiring trust. Gartner now predicts one in four job candidate profiles worldwide will be fake by 2028, and separate research shows most companies that promised to hire on skill alone never actually changed who they hire. DeweyLearn is betting that grading the work itself is the one signal neither problem can touch.

Inside DeweyLearn’s Oversubscribed $5 Million Round

DeweyLearn said it secured $5 million in an oversubscribed Series A round on July 16, with Catalysis Capital, Morningside and Owl Ventures joining lead investor SJF Ventures. SJF Ventures, founded in 1999, calls itself an impact venture fund built around businesses that create lasting, positive change.

Human expertise has been a limited resource for our entire history.

Luyen Chou, DeweyLearn’s co-founder and chief executive, said that in describing why the company built a machine to stand in for the instructors and experts who normally have to watch someone work in person. The company, based in New York, says its AI can observe an aspiring chef’s knife work or a nursing student’s bedside technique and return feedback modeled on what a master practitioner would say.

At the Auguste Escoffier School of Culinary Arts, the system has already graded more than 20,000 student submissions, giving instructors a first pass and getting feedback back to students faster than a human-only process allowed. Arrun Kapoor of SJF Ventures pointed to better learning outcomes and wider career access as the investment case, citing the founders’ combined background in industry and AI. The new money is earmarked to push the same approach into clinical and healthcare education, higher education, workforce learning and K-12 classrooms.

Two Hiring Signals Are Failing at Once

Start with the candidate side. Gartner, the research and advisory firm, predicts one in four candidate profiles will be fake by 2028. In a 2Q25 survey of 3,000 job candidates, 6% admitted to interview fraud, either posing as someone else or having a stand-in interview for them.

Trust runs the other way too. Only 26% of candidates believe AI will evaluate them fairly, even though 52% assume AI is already screening what they submit, Gartner found.

Greenhouse’s 2025 AI in Hiring Report, based on more than 4,100 job seekers, recruiters and hiring managers across the United States, United Kingdom, Ireland and Germany, found 74% of hiring managers are more worried about fake credentials and deepfakes than they were a year earlier. Nearly half of job seekers, 46%, say their trust in hiring has dropped in the past year, and 42% blame AI directly.

“Our latest data shows that neither side is happy with the hiring process right now,” Daniel Chait, Greenhouse’s chief executive, said in a statement.

Deepfake tools have made the fraud side easier to pull off. “Deepfake candidates are infiltrating the job market at a crazy, unprecedented rate,” said Vijay Balasubramaniyan, chief executive of voice authentication firm Pindrop Security, who has caught one himself.

The Skills-Based Hiring Promise Nobody Kept

The other signal breaking down is newer: the promise that skill would replace the degree. The National Association of Colleges and Employers found in its Job Outlook 2026 report that 70% of employers now use skills-based hiring for entry-level roles, up from 65% a year earlier. A TestGorilla survey put the figure even higher, with 85% of employers claiming to use some form of skills-based hiring.

Harvard Business School and the Burning Glass Institute, a labor-market research group, studied companies that publicly dropped degree requirements. They found 45% of those firms changed the posting but not the hiring, with fewer than 1 in 700 new hires actually landing without a college degree at some large employers. Real change was concentrated in the 37% of companies that rebuilt how they actually screen candidates rather than just editing a job listing.

The gap between policy and practice is exactly what a tool like DeweyLearn’s is built to close.

How Does the AI Grade a Knife Cut?

DeweyLearn’s system reads video, audio and other data captured while someone performs an actual task, then compares the steps against a standard set by the institution or employer beforehand. It is built to replace the instructor or expert who would otherwise have to watch every attempt in person.

The company calls it multimodal AI because it reads more than one kind of input at once, hand movement on video, what someone says while working, and the context of the task itself, rather than logging a multiple-choice answer or a checked box. Its models are trained on an institution’s own curriculum and domain expertise before they ever grade a student, so a nursing program’s rubric for inserting an IV looks nothing like a culinary school’s rubric for holding a chef’s knife.

Chou has described the payoff in specific terms: the system can walk an aspiring chef through her knife technique with feedback modeled on expert chefs, judge how effectively a clinical therapist runs a session, and let nursing students practice inside a simulated hospital before they ever touch a real patient.

A Skills Test Is Harder to Fake Than a Zoom Call

This is where the two problems intersect. A deepfake can carry a cloned face and voice through a video interview. It is much harder to fake finishing a real task end to end, on camera, in real time.

Gartner’s own research backs that instinct. 62% of candidates said they would be more likely to apply to a job that required an in-person interview, even with the extra friction that step adds to the process.

• Timestamped, one-sitting work – a task finished in a single monitored session leaves no window for a stand-in to finish it later.
• A real tool in the candidate’s hands – a knife, a live codebase, an actual customer call script makes it harder to fake competence from a script.
• Follow-up questions about a choice made minutes earlier – asking someone to explain a decision they just made catches a person who did not really do the task.

A small team does not need enterprise software to use all three. Building them into a single ninety-minute exercise, scored against criteria written down before anyone applied, works just as well.

Where Trust in AI Grading Still Breaks Down

DeweyLearn’s pitch assumes employers and candidates will both trust a machine’s judgment more than they trust a resume. Whether either side actually trusts that judgment yet is still unresolved.

Survey data pulls in different directions depending on who is answering.

• Employers, per a Resume Genius survey of 1,500 U.S. hiring managers: 87% say their organization already uses AI somewhere in hiring, and 65% believe it can help reduce bias.
• Candidates, per Greenhouse’s 2025 report: 35% think AI has simply moved bias from people to algorithms, and only 8% believe AI makes hiring more fair.
• Independent testing, per Brookings: AI hiring tools favored resumes with men’s names over women’s names 51.9% to 11.1% in controlled trials, and white-associated names over Black-associated names by a similarly wide margin.

DeweyLearn grades a finished task rather than screening an application, which sidesteps some of that resume-stage bias risk. It still faces the same basic question every AI grading system faces, whether the people being judged believe the judgment was fair.

Gartner’s own timeline runs to 2028. DeweyLearn is betting five million dollars that grading the work itself will still mean something by then.

Frequently Asked Questions

What does multimodal AI mean in DeweyLearn’s system?

It means the software reads more than one kind of input at the same time, video of how a task is performed, audio of what someone says while doing it, and the context of the assignment, then compares all of it against a standard the institution set in advance. That differs from older tools that only logged a multiple-choice answer or a click.

How much has candidate trust in the job search actually fallen?

One clear marker: Gartner found only 51% of candidates accepted a job offer in their most recent search as of the second quarter of 2025, down from 74% two years earlier. That decline tracks alongside rising candidate anxiety about deepfakes, AI screening and whether a posted job is even real.

Are companies actually hiring more people without a college degree?

The share of job postings requiring a bachelor’s degree has actually risen since 2024, according to Indeed’s Hiring Lab data on degree requirements, even as employers talk up skills-based hiring. That suggests the requirement is creeping back into parts of the market rather than disappearing.

Does AI hiring assessment actually reduce bias or add a new kind?

The evidence is mixed. One academic review found AI has not democratized entry-level hiring the way its backers promised, since success for young job seekers still tracks closely with family income and referrals rather than algorithmic scoring. That is a separate problem from the deepfake and credential fraud DeweyLearn’s category is built to catch.

How much money has DeweyLearn raised in total?

The $5 million Series A announced in July 2026 is DeweyLearn’s only disclosed funding round to date, according to startup funding trackers. The company has not disclosed a prior seed round or additional capital beyond what SJF Ventures and its co-investors put into this raise.