Kintsugi, a Berkeley-based startup that built AI technology to detect depression and anxiety from voice patterns, has shut down commercial operations after failing to secure FDA clearance. The company, which raised $30 million and spent four years pursuing regulatory approval, exhausted its funding before completing its De Novo submission to the FDA.
The startup's technology, called KiVA (Kintsugi Voice Biomarker API), analyzed 20 seconds of free-form speech to identify vocal signs of depression and anxiety. Unlike traditional screening methods that rely on questionnaires, the system detected subtle acoustic features like pitch, tone, energy, rhythm, and pauses that correlate with mental health states. The approach was language-agnostic, working across any spoken language.
Kintsugi validated its technology in a prospective, double-blind pivotal study comparing its model against SCID-5 clinician interviews. The results demonstrated performance that exceeded the roughly 47% detection rate typical in primary care settings. The company claimed over 80% accuracy compared to clinician assessments, earning a 2022 Frost & Sullivan innovation award.
Despite promising results and market interest, the startup couldn't survive the regulatory timeline. Founder Grace Chang invested $16 million in regulatory efforts alone, pursuing FDA clearance while clinical buyers refused to purchase without it. The lengthy FDA review process created what industry observers call a "regulatory dead zone" where startup runways couldn't match multi-year approval timelines.
In February 2026, Kintsugi ceased operations and released all AI models, methodologies, and research as open-source on Hugging Face. Chang stated the company chose "the integrity of the science over the limitations of a distressed market." The shutdown highlights a persistent challenge in health AI: companies with validated technology and clear demand still struggle to navigate the FDA's costly and time-intensive clearance process.
The timing is particularly troubling given mental health crisis rates. Only 4% of U.S. primary care visits currently screen for depression, despite recommendations from medical bodies. Voice-based screening could fill that gap at scale, but the economics of FDA approval remain prohibitive for early-stage companies.