“We don’t want to capture the static person,” co-founder and CEO Abhivyakti Ahuja said. “We want to capture the changing person. That means capturing the longitudinal data on how people are changing, what triggers are changing them and to what degree.”
The model combines client customer data with signals including social media, pop culture, and current events. Changes in behavior are treated as useful data themselves, while a lack of change can also help the system determine how a population is responding to outside influences.
Mirror Particle is positioning that approach as an alternative to systems that use large language models to role-play as particular types of consumers. Ahuja argues that written language alone does not capture all the information people use when making decisions.
“LLMs are modeling written language, but humans are made of visual perception, spatial reasoning, social intelligence,” she said.
The company is also placing more emphasis on what it calls revealed behavior — what people actually do — instead of depending primarily on surveys or interviews about what consumers say they intend to do.
That distinction shapes how Mirror Particle expects companies to use the technology. Rather than simply testing advertising language for a target audience, a business could use the system to examine whether the product itself is suited to that group.
For example, Ahuja said a beauty company targeting younger consumers could use behavioral modeling to determine whether customers want the proposed product category at all.
“What if the target demographic doesn’t want eyeshadow palettes?” Ahuja said. “Maybe blush is a better option to go for if you want to sell a product to this market.”
Mirror Particle is initially pursuing customers with existing budgets for consumer insights, including research teams and groups responsible for product and brand strategy. The company says its prediction engine is intended not only to forecast behavior but also to explain the motivations, limitations, and circumstances behind its conclusions.
In one early pilot, a pet food company wanted to determine which type of imagery should appear on its packaging. Mirror Particle’s analysis concluded that the imagery was not the main issue. Instead, the company said the brand’s strong recognition had contributed to a mass-market and inexpensive perception that could limit sales.
That focus on underlying causes is meant to make the system useful for business decisions where a prediction alone may not provide enough information. Teams can use the accompanying explanation to evaluate why a model expects a particular outcome and what factors could be influencing it.
Mirror Particle describes its broader goal as building a model of human behavior rather than another system centered primarily on generating language. Ahuja compares the company’s development approach to the way people gradually learn about their surroundings through different forms of perception and interaction.
“The way we see our model evolving is like how a baby learns about the world,” Ahuja said.
Her interest in the problem draws on a background in neuroscience and computer science. She later worked at Amazon Robotics, where she met co-founders Will Song and Thomson Yen. Song has worked on sales personalization technology, while Yen has focused on deep learning and how AI agents interpret human behavior.
The two-year-old company has raised an angel round and said it is close to completing its first venture round. It is also among the startups competing in Startup Battlefield 200 at TechCrunch Disrupt 2026.
Mirror Particle ultimately wants to move beyond broad demographic modeling toward more detailed predictions about individuals. The company describes that ambition as creating a general layer for anticipating human behavior.
“We just need a better model of humans if we’re going to work alongside AI and with each other,” Ahuja said.
This analysis is based on reporting from TechCrunch & Konsulteer.
Image courtesy of Dealroom.
This article was generated with AI assistance and reviewed for accuracy and quality.
