Mirror Particle builds a ‘model of the world’ of human behaviour rather than relying on an LLM

This San Francisco-based start-up aims to predict what consumers actually do, and why. It is targeting a niche in which Simile, Aaru and Humans& have already raised hundreds of millions.

Created two years ago in San Francisco, Mirror Particle is developing a foundation model to anticipate consumer behavior and the motivations behind it. Its CEO, Abhivyakti Ahuja, has a dual background in neuroscience and computer science, acquired at the University of Toronto in the wake of Geoffrey Hinton. She co-founded the company with Will Song, a specialist in commercial personalization, and Thomson Yen, who came from deep learning; all three worked together at Amazon Robotics.

Observed behaviour rather than questionnaires

The team challenges the approach of competitors who tune large language models to simulate consumers: according to them, some refinement data carries little weight compared to the hundreds of billions of examples of LLM training. Mirror Particle therefore builds its own model, fed by customer data, news, popular culture and social media trends. It focuses on what people actually do rather than what they say, and tracks changes in behavior over time.

The company first targets market research and brand strategy. Case in point: a pet food brand realized that its packaging was not to blame, but that its image as a consumer product was holding back its sales. After a round of angel investors, Mirror Particle is preparing to close its first round of fundraising, in a market where Simile ($200 million), Aaru ($88 million) and Humans& ($480 million in seed funding) have already attracted a lot of capital. It is participating in TechCrunch Disrupt's Startup Battlefield, from October 13 to 15.