TechCrunch Reporter Tests Synthesia's Interactive Digital Twin Technology

London-founded AI video startup Synthesia has expanded its interactive avatar technology by creating a digital twin of TechCrunch senior venture capital reporter Dominic-Madori Davis. The project marks the first instance of Synthesia crafting a custom interactive avatar for a journalist, showcasing how synthetic media tools are shifting from basic script-reading videos into conversational agents.

The creation process took place at Synthesia's new office space in New York City. To construct the digital clone, Davis entered an onsite mini film studio where the company captured multiple photographs and recorded a two-minute voice sample following explicit consent. Synthesia subsequently generated two distinct variations of the reporter's digital twin: a personal avatar designed to read user-provided scripts, and an interactive avatar engineered to process live audio queries and respond in real time.

The Tech Stack Behind Conversational Avatars

Synthesia's interactive avatar platform operates on a multi-layered model architecture. When a user speaks to the digital twin, a voice-to-text model transcribes the audio into text. An agentic language model then interprets the text and determines appropriate actions or answers. Next, a text-to-voice model converts the system's written response into spoken audio, while Synthesia's proprietary video model animates the visual avatar to match the vocal output.

Although Synthesia offers its own proprietary video and audio models, the startup permits enterprise clients to integrate alternative components from external providers, including OpenAI, Google, ElevenLabs, or Cartesia. Enterprise customers can also choose to host their avatars on Synthesia's infrastructure or deploy them within their own private cloud environments.

Deterministic Guardrails and Initial Testing

To evaluate the interactive avatar under controlled conditions, Synthesia configured Davis's digital twin as a deterministic model. Instead of functioning as an open-ended conversational bot, the interactive avatar was trained strictly on a single article authored by Davis regarding why venture-backed startups commit fraud at higher rates than non-funded counterparts.

During practical tests detailed by TechCrunch, the avatar strictly adhered to its assigned knowledge base. When asked personal background questions—such as where the reporter lived or worked prior to joining TechCrunch—the avatar refused to answer non-related topics and directed users back to the venture fraud article. Friends and family testing the avatar noted that while the standard script-reading version reproduced her voice accurately, the interactive twin felt somewhat uncanny while remaining strictly bounded by its training data.

Enterprise Scaling and Market Context

Synthesia's media experiment follows a period of rapid commercial expansion for the startup. Synthesia hit a $4 billion valuation earlier this year after reporting that it crossed $100 million in annual recurring revenue last year. Operating in an increasingly competitive avatar market alongside startups like HeyGen, D-ID, and Colossyan, the firm has broadened its core offerings beyond simple video production.

The enterprise platform currently centers on three main product lines: a standard video creation tool where avatars read typed scripts; an agentic interaction platform named Sessions, which includes products like Roleplay Sessions for sales training and practice; and an API suite that allows developers to integrate Synthesia's voice and video models into third-party software.

Why It Matters

The development of deterministic digital twins highlights a strategic shift in enterprise AI deployment. By constraining an avatar's responses strictly to pre-approved corporate datasets or specific investigative reports, companies can effectively prevent AI hallucinations and unscripted errors. This controlled interaction model offers immediate utility for enterprise applications, such as automated employee roleplay, training modules, and routine client inquiries, allowing individuals to delegate specific information-sharing tasks without active oversight.

At the same time, bringing interactive clones into professional fields like journalism accentuates ongoing questions surrounding public trust and authenticity. While automated avatars can handle routine script readings or answer static inquiries about published articles, primary newsgathering and critical analysis rely fundamentally on human credibility. As synthetic media becomes more prevalent across digital platforms, enterprises and media organizations face the ongoing challenge of balancing the efficiency of interactive avatars against potential public resistance to AI-generated representatives.