How did Chandra De Keyser build the proprietary data set for Voicera's sincerity AI model?
Because no existing sincerity data set was available, Voicera built its own from scratch — deliberately. The team designed the collection process to include a diverse mix of ethnicities, cultures, and languages, drawing contributors from countries including China, India, and South Africa, specifically to avoid building a model trained on a whites-only bias. Every step was governed by full legal and ethical guidelines, with genuine informed consent rather than terms buried in a 10-page document contributors would never read.
De Keyser made an important distinction in his approach to consent: contributors had to genuinely understand they were providing video footage for AI training purposes. Real lawyers were involved throughout the entire process — not as an afterthought, but as an integral part of building the data set. This level of care reflects a broader philosophy at Voicera: that understanding human sincerity at scale requires starting with data that is itself collected with integrity.
Sincerity AI is the term Voicera uses to describe technology designed to help organizations understand not just what customers say, but what they mean — by combining large language models with emotional intelligence signals captured from voice and video. It is distinct from generative AI, which primarily produces content such as text, images, or video.
The decision to build a proprietary data set rather than rely on existing sources was not merely a technical one. It was a product and ethical decision. Widely available data sets in the emotional AI space have historically skewed heavily toward Western, predominantly white populations — a known bias that would have undermined the core promise of sincerity AI: to understand humans across cultures and contexts. De Keyser's previous company, MoodMe, gave him over a decade of experience navigating exactly these challenges in emotional AI and facial analysis, which makes his approach to data collection on this episode particularly credible. You can follow the full conversation on The Conference Room on Listenly.
"What was said is part of the story. How it was said is sometimes as important, if not more important. So that's where we add tremendous value, because we can unherd these signals."
— Chandra De Keyser, Co-founder, VoiceraAbout Chandra De Keyser
Chandra De Keyser is the co-founder of Voicera, an AI company pioneering sincerity AI technology designed to help organizations understand not just what customers say, but what they truly mean — by combining large language models with emotional intelligence. Voicera's technology is already deployed at scale: one client operates a car dealer network of approximately 200 dealerships across the US, handling hundreds of calls per day through Voicera's voice agent.
Before Voicera, De Keyser founded MoodMe roughly 14 years ago, making it one of the earliest innovators in emotional AI and facial analysis — years before generative AI became mainstream. MoodMe grew to approximately 160 customers in the augmented reality space, including a deployment for the FIFA Women's World Cup around seven to eight years ago, before pivoting toward emotion detection.
His career spans Europe, Silicon Valley, and Latin America. He began with a master's degree in computer science in Italy, then joined a startup in the 1980s working on an early touchscreen device from the 83rd floor of the World Trade Center. This deep, decades-long track record in human-centred AI is precisely what makes De Keyser's deliberate, ethics-first approach to data collection credible — and worth listening to in full.
See also
Generative AI is a broad category of AI that generates content — mainly text through LLMs like ChatGPT, Claude, or Grok — as well as images and video. Sincerity AI goes further by analyzing not just what is said, but how it is said, layering emotional intelligence on top of language understanding.
As of the time of the interview, Tracefuse works with 700 brands and has removed over 16,000 reviews. Shane Barker attributes this growth primarily to the quality and rigor of their filing methodology.
Barker states that if you compared Tracefuse's filing methods from five and a half years ago to today, they are "almost polar opposites," largely because the team continuously refined their approach based on platform feedback and evolving guidelines.