Mirror Particle Is Building AI That Predicts Human Behavior

“Mirror Particle” is developing an artificial intelligence model that predicts people’s future behavior and the reasons behind their decisions. The 2-year-old San Francisco-based startup calls the system a “world model” built from scratch. The model learns about the environments people live in and their reactions to those environments together. According to the company, the system will also track how consumer behavior changes over time.

Interest in this field has grown rapidly over the past 1 year. “Simile” was valued at $2 billion (approximately 3.4 billion manats) and raised $200 million (approximately 340 million manats). “Aaru” raised $88 million (approximately 149.6 million manats) at a valuation of $1 billion (approximately 1.7 billion manats). In January 2026, “Humans&” announced its first $480 million funding round (approximately 816 million manats). At the time, the company was valued at $4.48 billion (approximately 7.62 billion manats) and launched a system called “Persimmon.”

“Mirror Particle” says its approach differs from systems that use large language models to create virtual replicas of target audiences. According to the company, models trained mainly on written text do not fully reflect how people perceive what they see, their spatial reasoning, or their social behavior. Additional training with a small amount of new data also does not sufficiently change biases arising from old data.

The model combines customer data with information collected from daily events, pop culture, and social media. The primary focus is not on what people say in surveys, but on how they actually behave. The system tracks which experiences cause people’s opinions to change and what influences those changes. When no change occurs, it also records that as a separate indicator.

Initially, the technology will be used for market research and brand and product strategy. For example, the system does not merely help a cosmetics brand write advertising copy tailored to Generation Z. It can also determine whether that audience actually wants the product. The prediction system does more than make recommendations; it also shows customers the desires, barriers, and other factors that influence people’s choices.

In one of the first trials, a well-known pet food brand wanted to learn which image it should choose for its packaging to increase sales. The model determined that the problem was not the image of chicken, beef, or vegetables. Because the brand was so well known, buyers perceived it as a mass-market, inexpensive product. Therefore, it was this perception of the product, rather than the packaging, that was preventing sales growth.

The startup has received initial funding from individual investors and announced that it is close to completing its first venture funding round. “Mirror Particle” will participate in the “Startup Battlefield 200” competition, which will be held in San Francisco on October 13-15, 2026. Abhivyakti Ahuja co-founded “Mirror Particle” with Will Song and Thomson Yen, whom Ahuja met at “Amazon Robotics.” The startup’s long-term goal is to move from analyzing population groups to predicting the behavior of individual people.