Synthesia built a TechCrunch reporter an interactive avatar

TechCrunch reporter Dominic-Madori Davis now has an interactive AI avatar of herself that answers spoken questions about one of her articles and refuses to talk about anything else. Synthesia made it after she visited its New York office in September. Davis writes that it is the first digital avatar the company has made for a journalist, or for anyone other than Alexandru Voica, its head of corporate affairs, whose own avatar answers press questions.

Davis wrote up the build on September 26. Synthesia's team took numerous photographs of her in a studio inside the office and recorded about two minutes of her voice, then took a couple of days to make her avatars: a personal one, which reads whatever script she gives it, and an interactive one, which listens and talks back, each made with and without glasses, four versions in all.

The interactive avatar is deterministic. It was trained on a single article, her July 31 story on research finding that venture-backed startups commit more fraud than startups without venture backing, and it steers every off-topic question back to that article. Davis asked it where she was before TechCrunch and which part of New York she lives in; it redirected both times. Her parents tried questions only they would know, and got the same redirect. "I don't remember giving birth to two of you," her mother said after testing it.

Davis describes the stack behind her avatar as a voice-to-text model, an agentic language model, a text-to-voice model and a video model, built by Synthesia, that animates the avatar as it talks. Hers uses Synthesia's own video and voice models; customers can choose alternatives from other labs such as Cartesia, ElevenLabs, Google or OpenAI, and Synthesia's interactive avatars page says teams can connect their own language model. Enterprises can host their avatars on whatever cloud they want or pay Synthesia to host them.

The company has three product lines: a video platform where an avatar reads a typed script, an agentic product called Sessions used for surveys and roleplay, and an API for combining its models with other services. Synthesia hit a $4 billion valuation in January 2026, and said in April 2025 that it had crossed $100 million in annual recurring revenue.

Davis's friends found the likeness on the interactive version weaker than on the personal one, and still "somewhat creepy." She reports that one investor, asked whether they would accept news presented by an avatar, said no immediately. Because her avatars are deterministic, Davis writes, they will never say anything new; she can see how a nondeterministic one, powered by a chatbot free to pontificate, could tip someone into "a tinge of AI psychosis."

The avatar is embedded in the TechCrunch article and can be questioned there: techcrunch.com