Mirror Particle pursues a changing model of consumer behavior
San Francisco startup Mirror Particle is building a foundation model intended to predict consumer behaviour and explain the motivations behind it. The two-year-old company says it has raised an angel round and is close to closing its first venture round, while preparing to compete in TechCrunch Startup Battlefield.
Its central claim is that the common approach to behavioural prediction—prompting or fine-tuning large language models to roleplay a demographic—does not adequately represent how people make decisions. Mirror Particle instead describes its product as a “world model” built from scratch to simulate why people act and how their behaviour changes over time.
From static segments to changing people
Co-founder and chief executive Abhivyakti Ahuja argues that language models are trained to model written language, whereas human experience also involves visual perception, spatial reasoning and social intelligence. In her view, an analysis designed around a static demographic can miss the triggers and experiences that alter motivations.
Mirror Particle aims to capture longitudinal signals: how people change, what is changing them, and by how much. A lack of change is also treated as a signal. The company models a demographic as an evolving system rather than as a fixed persona, with an initial focus on market research and brand and product strategy.
Data sources and revealed behaviour
The startup says its proprietary data combination includes customer data from clients, current events, pop culture and social media. Much of its attention is directed to “revealed behaviour”—what people actually do—rather than to self-reported answers in surveys.
That distinction shapes the product’s proposed use. A beauty brand, for example, could use the engine not only to refine advertising copy for Gen Z but also to assess whether the group wants a particular product category at all. Ahuja suggested that a company considering eyeshadow palettes might find that blush is the more relevant product opportunity.
An explanation alongside the prediction
Mirror Particle says its engine provides customers with the reasoning behind a recommendation, including motivations, constraints and additional context. In an early pilot with a well-known pet-food brand, the client asked which packaging image—such as chicken, beef or vegetables—would best lift sales.
The system’s conclusion was that imagery was not the decisive issue. Mirror Particle found that the brand was seen as mass market and cheap because it was so recognisable, and that sales would plateau until that perception was addressed. The example illustrates the company’s effort to challenge the premise of a research question rather than merely select among supplied creative options.
Ahuja studied neuroscience and computer science at the University of Toronto before working at Amazon Robotics, where she met co-founders Will Song and Thomson Yen. Song has worked on sales-personalisation engines, while Yen focused on deep learning and how AI agents understand human behaviour. For businesses, the immediate implication is to treat behavioural AI as a tool for testing product and positioning assumptions, not simply for generating more tailored campaign language.