Kintsugi
The business, the turning points, and the decisions behind them.
01 / WHAT IT BUILT
What the company did
Kintsugi was a California AI startup founded in 2019 by Grace Chang and Rima Seiilova-Olson. It built voice biomarker technology that could detect clinical depression and anxiety from 20 seconds of speech by analyzing vocal features like pitch variation, speaking rate, pauses and timbre. It assembled what was described as the world's largest annotated voice repository for mental health, with peer-reviewed results published in the Annals of Family Medicine, and had commercial relationships with major health payers and providers.
02 / Kintsugi FAILURE REASONS
Failure factors
Kintsugi shut commercial operations after failing to secure FDA De Novo clearance in time; it ran out of funding while waiting for its final FDA submission, additional fundraising faltered as runway shortened, and the team rejected predatory short-term financing offers before open-sourcing most of the tech.
the company failed to secure FDA clearance in time while pursuing the De Novo pathway for its voice biomarker AI.
Chang said the startup ran out of funding waiting for its final FDA submission.
Efforts to raise additional funds faltered as runway shortened; the team declined a predatory offer (~$50k/week for $1M equity) rather than meet payroll that way.
03 / EVENT HISTORY
Events in the structured record
- Operations ended
company shutting down commercial operations and open-sourcing voice biomarker models/research; cites a distressed market and refusal of closed acquisition. Corroborated by Behavioral Health Business (11 Feb 2026) and The Verge (2 Apr 2026).
04 / CONTEMPORARY REBUILD ASSESSMENT
Difficulty of rebuilding this project today
Material for this section is being gathered and checked. It will appear here once ready.
Overall assessment: Not completed
05 / CAUSAL CHAIN
How the factors connect
- 01
In early 2026 Kintsugi announced it was ceasing commercial operations after seven years and about $28 million in funding.
- 02
The company had spent about $16 million over four years pursuing FDA clearance for its depression/anxiety detection technology and failed to secure it in time.
- 03
Rather than continue down an uncertain regulatory path, leadership decided to wind down the company.
- 04
It released its core technology as open-source software, including its Depression-Anxiety Model trained on roughly 863 hours of speech from 35,000 individuals; the raw audio data remains private.