BREAKING
Grace Chang - Founder & CEO, Kintsugi    $28M raised to democratize mental healthcare through voice AI Inc. Female Founders 500 (2025) Forbes AI 50 North America Gartner Cool Vendor for Responsible AI Fierce Healthcare Fierce 15 (2023) 80% accuracy detecting depression from 20 seconds of voice Kintsugi technology released to public domain Feb 2026 Japan's default mental health screener World's largest annotated voice dataset for mental health ML        Grace Chang - Founder & CEO, Kintsugi    $28M raised to democratize mental healthcare through voice AI Inc. Female Founders 500 (2025) Forbes AI 50 North America Gartner Cool Vendor for Responsible AI Fierce Healthcare Fierce 15 (2023) 80% accuracy detecting depression from 20 seconds of voice Kintsugi technology released to public domain Feb 2026 Japan's default mental health screener World's largest annotated voice dataset for mental health ML
Founder & CEO  |  Kintsugi Health  |  Berkeley, CA

Grace
Chang

Voice. Gold. Repair. The algorithm that listened to America.
She built an AI that hears what you cannot say - then gave it to the world for free.
$28M
Total Raised
80%
Detection Accuracy
20s
Of Speech Needed
5x
Serial Founder
Grace Chang, Founder & CEO of Kintsugi
Photo: Kintsugi Media Kit
Inc. Female Founders 500 (2025)
Forbes AI 50 North America
Gartner Cool Vendor - Responsible AI
Fierce Healthcare Fierce 15 (2023)
NSF Phase I & II Awardee
ODF Top 1% (2025)
01

Who She Is

Twenty seconds. That is all Grace Chang's AI needs. Not a questionnaire. Not a clinical intake form. Not the forty-five minutes of a $450-an-hour therapy session she once had to pay out-of-pocket because she could not find care for six months. Just twenty seconds of your voice - talking about anything, in any language - and Kintsugi's model would tell you, with 80% clinical accuracy, whether you were carrying depression or anxiety.

Chang arrived in the United States from Taiwan as a child, one of the immigrants who came with modest means and rebuilt from scratch. She studied computer science at USC on scholarship in 2000, joining a cohort of 230 students where she was one of two women. She described it plainly: the ratio did not improve much as she climbed through Silicon Valley over the next three decades. Some executives told her to "stay in your lane and act more subservient." She built companies instead.

By 2018, she was Head of Product at UnifyID, an authentication startup that launched at TechCrunch Disrupt and became a runner-up in Startup Battlefield, won at South by Southwest, and raised a $20 million Series A with just five people on the team. Chang had a knack for signal processing - the math of extracting meaning from noise. She did not know yet that she was about to apply it to a different kind of signal entirely.

The pivot came from burnout and a trip to Japan. After struggling for six months to access mental health care and ultimately paying $450 an hour for therapy, she recognized the gap between need and access was not a rounding error - it was structural. In Japan, she encountered the ancient art of kintsugi: broken ceramics repaired with gold, the cracks made visible rather than hidden, the object more beautiful for having been broken. She came home with the name of a company she had not yet started.

At an open AI hackathon in San Francisco in 2019, she met Rima Seiilova-Olson - a machine learning scientist who had grown up in a yurt in Kazakhstan without running water or electricity. They were a startling pair: one raised in Taiwan, the other in Central Asia; one an engineer-entrepreneur, the other a researcher. Together, they built Kintsugi. The company received an NSF Phase I grant in October 2019 and never looked back.

From the very early days, I was interested in understanding how things were put together from the ground up.

- Grace Chang

What Kintsugi built was genuinely strange and important: an AI that listened to how you spoke, not what you said. The model analyzed vocal biomarkers - pitch, cadence, timbre, micro-pauses - patterns that correlate with clinical depression and anxiety. Because it did not process language content, it was language-agnostic. It worked in Mandarin, Spanish, Japanese, English. It worked across gender and socioeconomic background. Chang insisted on building the world's largest annotated voice dataset for mental health machine learning - 110 times larger than competitors - specifically to avoid the algorithmic biases that plague healthcare AI built on unrepresentative training data.

The platform, Kintsugi Voice, found its way into clinical call centers processing 20 million calls annually. Kintsugi was selected as one of 10 companies from 2,600 applicants to work with UnitedHealth Group. In 2024, Kintsugi became Japan's default mental health screener. A peer-reviewed paper appeared in the Annals of Family Medicine in January 2025. Kintsugi hit a 5000% growth in patient screenings over one year. Seven new partnerships. Zero customer churn. A perfect NPS of 10.

Then, in February 2026, Grace Chang announced that Kintsugi would cease commercial operations and release all of its research, models, and technology into the public domain. The reason was not failure - it was physics. FDA regulatory pathways for AI-based diagnostics are expensive and slow. A venture-backed business model could not survive the cost and timeline. Rather than watch the technology sit behind a paywall or quietly dissolve, Chang chose the harder thing: open source everything, let the science live in the commons, give the work a future outside the limits of any single company.

It was a kintsugi move. The company cracked. The gold is visible now.

02

The Technology

20s
Free-Form Speech
The model needs no scripted prompts, no clinical context. Any twenty seconds of natural speech in any language is sufficient input for a clinical-grade assessment.
80%
Clinical Accuracy
Kintsugi Voice detects clinical depression and anxiety at 80%+ accuracy - matching or exceeding many conventional screening tools while operating passively in existing call flows.
110x
Dataset Scale
Chang built the world's largest annotated voice repository for mental health machine learning - 110 times larger than the nearest competitor - specifically to reduce demographic bias.
Detection Performance
Depression Detection80%+
Anxiety Detection80%+
Language Agnostic CoverageAll languages
Dataset Lead vs. Competitors110x

Source: Kintsugi clinical validation, Annals of Family Medicine 2025

How It Works
1
Patient speaks naturally - in any language - during a call center interaction, telehealth visit, or through the app
2
AI analyzes vocal biomarkers - pitch, cadence, timbre, micro-pauses - not the words themselves
3
Clinical flags surface in real time, enabling prioritization and care routing without patient self-report
+
Side discovery: models detect synthetic/deepfake voices because AI-generated speech lacks biological and cognitive vocal attributes
03

The Origin Story

From Signal Processing to Suffering - and Back

There is a through-line in Grace Chang's career that is easy to miss if you are looking at job titles. Before Kintsugi, she spent years at startups that extracted meaning from complex signals. At UnifyID, that meant behavioral patterns - the way you walk, type, or hold your phone - synthesized into an authentication system that could recognize you without passwords. The work was elegant: instead of something you know or something you have, identity became something you are, encoded in motion and habit.

Voice is a more intimate signal. It carries emotion, fatigue, cognitive load. When Chang finally found a therapist after six months of trying and $450 an hour of expense, she was not just solving her own problem. She was documenting a failure mode in the healthcare system: the gap between need and access was widest precisely for people under the most stress.

Japan offered the metaphor. Kintsugi - the art of repairing broken pottery with gold - is not a practice of concealment. The repair is made beautiful. The crack becomes the story. Chang took the name because it matched the product: a tool that did not hide distress but illuminated it early, so it could be addressed before it shattered.

The Hackathon

"Met Rima Seiilova-Olson at an open AI hackathon in San Francisco. One from Taiwan. One from a yurt in Kazakhstan. Both without access to mental health care. They built Kintsugi in 2019."

The Accidental Discovery

"Kintsugi's mental health voice AI turned out to detect deepfakes too. Synthetic voices don't carry the same biological and cognitive attributes as real human speech. The model knew the difference every time."

The Open Source Decision

"In February 2026, rather than let FDA regulatory costs kill the research, Chang released everything to the public domain. Seven years of voice biomarker science. Free. Forever."

04

Career Timeline

c. 2000
Enrolled at USC on scholarship - Computer Science and Economics. One of two women in a cohort of 230.
Early Career
Three decades building technology across security, enterprise, consumer, health, and fintech companies in the Bay Area. Five companies total.
2018
Head of Product at UnifyID. Launched at TechCrunch Disrupt; runner-up Startup Battlefield; won SXSW. Raised $20M Series A with five-person team.
2019
Co-founded Kintsugi with Rima Seiilova-Olson at an AI hackathon. NSF Phase I grant awarded October 2019. Launched Apple App Store's first voice journaling app.
2021
Joined On Deck Founders ODF5 cohort during COVID-19 pandemic. Referred co-founder Rima to ODF6. Closed seed round on New Year's Eve.
Feb 2022
$20M Series A from Insight Partners. Total capital: $28M. Investors include Acrew Capital, Darling Ventures, Citta Capital, Alpha Edison.
2023
Fierce Healthcare Fierce 15. Forbes AI 50 North America. Gartner Cool Vendor for Responsible AI. Milken Institute Future Health Summit speaker.
2024
Kintsugi becomes Japan's default mental health screener. Launches Kintsugi Signal for deepfake detection. 5000% patient screening growth.
Jan 2025
Peer-reviewed research published in Annals of Family Medicine. ODF Top 1% recognition. Inc. Female Founders 500 List. Perfect NPS 10/10 across 7 new partnerships.
Feb 2026
Kintsugi ceases commercial operations. All AI research, models, and technology released to the public domain. FDA regulatory burden cited as unsustainable for venture model.
05

In Her Words

"From the very early days, I was interested in understanding how things were put together from the ground up."
Grace Chang - on her lifelong drive to build
"We have been able to identify voice biomarkers as it correlates to depression through short snippets of free form speech."
Grace Chang - on Kintsugi's core technology
"We aim to provide objective, quantifiable, and accurate measurements to raise the parity of mental health to that of physical health."
Grace Chang - Kintsugi Series A announcement, 2022
"I'm truly humbled to be recognized as one of Inc.'s top female founders for 2025. At Kintsugi, we're determined to break down barriers to mental health care access."
Grace Chang - Inc. Female Founders 500, March 2025
06

Recognition & Achievements

Inc. Female Founders 500 List (2025) - recognized for authentic leadership as an Asian female founder in healthcare AI
AI
Forbes AI 50 for North America - one of the continent's most influential AI companies
G
Gartner Cool Vendor for Responsible AI - recognized for bias mitigation and unbiased data collection practices
F
Fierce Healthcare Fierce 15 (2023) - top 15 most disruptive companies in digital health
$
$28M raised from Insight Partners, Acrew Capital, Darling Ventures, and others. NSF Phase I & II SBIR grants.
JP
Kintsugi selected as Japan's default national mental health screener - first major international government deployment
UN
1 of 10 companies selected from 2,600 applicants to partner with UnitedHealth Group - integrated into 20M-call clinical pipeline
📄
Peer-reviewed research published in Annals of Family Medicine (January 2025) on voice AI in clinical settings
ODF
On Deck Founders Top 1% recognition (November 2025) - among the most impactful companies in the ODF portfolio
07

Watch Grace Chang

Featured Interview

Can AI Tell How You're Feeling?

Grace Chang discusses Kintsugi's voice biomarker software and how it detects clinical signs of depression and anxiety - without listening to what you say, only how you say it.

A clear, accessible explanation of the technology from the person who built it: the signal processing logic, the clinical validation process, and the ambition to make mental health screening as routine as a blood pressure reading.

08

The Open Source Decision

When the company ended, the science began.

In February 2026, Grace Chang made a decision that few founders make: instead of shutting Kintsugi down quietly, she released everything - seven years of research, the AI models, the voice biomarker dataset, the clinical validation work - to the public domain.

The reason was not defeat. It was economics colliding with ambition. FDA regulatory pathways for AI-based diagnostics cost millions and take years. A venture-backed company, dependent on growth timelines and return horizons, could not survive that timeline. The technology itself was validated. The business model was not.

Chang chose permanence over closure. The research is now available to anyone: researchers, nonprofits, healthcare systems, governments. The world's largest annotated voice dataset for mental health machine learning, open and free.

What Was Released
All AI models, research papers, the annotated voice dataset (110x larger than competitors), and clinical validation frameworks
Why It Matters
FDA regulatory costs made the venture model unsustainable - but the science was too important to let die with the company
What Comes Next
Kintsugi's voice biomarker research is now a foundation for academic institutions, health systems, and governments globally
Unintended Legacy
The deepfake detection capability of Kintsugi Signal - an accidental discovery - also enters the public domain
09

Fun Facts

🧋
Daily treat: hot hojicha boba from Boba Guys with oat milk. Non-negotiable.
🎊
Closed Kintsugi's Series A on New Year's Eve 2021. Champagne had dual purpose.
🎙️
She uses Kintsugi's own voice journaling app every day to track her personal stress triggers.
🕵️
Kintsugi accidentally invented deepfake detection - mental health AI that can spot synthetic voices.
🌍
The AI works in every language - because it listens to how you speak, not what you say.
🛖
Her co-founder grew up in a yurt in Kazakhstan with no running water. They met at a hackathon.
📱
Daily screen time: 4 hours 48 minutes. Primary apps: Email, Slack, Safari, YouTube, TikTok.
💛
Gives employees $150/month wellness stipends and Summer Fridays - because she lived the burnout.