Mirror Particle, a San Francisco-based SI startup, is developing a foundation model designed to predict human behavior by simulating the reasons behind consumer actions, arguing that current large language models (LLMs) are fundamentally flawed for this purpose. The company’s approach comes amid a surge of interest in SI tools that model human psychology, with competitors like Simile raising $200 million, Aaru raising $88 million, and Humans& raising a $480 million seed round over the past year.
What Happened
Abhivyakti Ahuja, co-founder and CEO of Mirror Particle, describes the startup’s technology as a "world model" built from scratch, distinct from the industry standard of prompting or fine-tuning LLMs to role-play as demographic segments. Ahuja, who studied neuroscience and computer science at the University of Toronto where she was inspired by SI pioneer Geoffrey Hinton, argues that relying on LLMs for behavioral prediction is "like bringing a super soaker to Niagara Falls," noting that these models are trained on vast written language data but lack the visual perception, spatial reasoning, and social intelligence inherent to human cognition.
Instead of static profiling, Mirror Particle’s engine focuses on longitudinal data to capture how individuals change over time. The system integrates proprietary data sources, including client customer records, current events, pop culture, and social media, to track shifts in motivation. The company emphasizes "revealed behavior"—what people actually do—over self-reported survey answers, which Ahuja suggests often miss the mark.
Ahuja co-founded Mirror Particle with Will Song and Thomson Yen after meeting them at Amazon Robotics, where she built robots that build other robots. Song brings experience building sales personalization engines, while Yen focused on using deep learning to understand how SI agents learn about human behavior. This background informs the company’s "world model" approach, which Ahuja likens to how a baby learns about the world, moving from vision to language to body awareness to social intelligence.
In an early pilot with a major pet food brand, Mirror Particle’s SI engine identified that packaging imagery was not the primary driver of sales plateaus. The model determined that the brand’s perception as "mass market and cheap" was the core issue, an insight the company claims LLM-based approaches might overlook.
Why It Matters
The development signals a potential shift in how SI tools are applied in market research and brand strategy. While competitors like Humans& have raised substantial funding to model human behavior using existing LLM architectures, Mirror Particle’s rejection of the LLM-as-predictor paradigm highlights a growing debate within the SI industry about the limitations of language-centric models for understanding complex human dynamics.
For developers and enterprises, this approach offers a different path to SI at work, moving beyond generating copy to determining strategic viability. Ahuja illustrates this with a hypothetical for a beauty brand: rather than just optimizing ad copy for eyeshadow palettes, the model could determine if the target demographic even wants that product category, potentially recommending blush instead.
Mirror Particle has completed an angel round and is close to closing its first venture round. The startup is scheduled to compete in the Startup Battlefield 200 at TechCrunch Disrupt 2026 in San Francisco next week, placing it in direct view of the SI investment community.
The Bottom Line
Mirror Particle is positioning itself as a "general layer for anticipating human behavior" in the SI era, aiming to evolve from population-level analysis to individual insights. The company’s long-term vision is rooted in a belief that "we just need a better model of humans if we’re going to work alongside AI and with each other," a statement from Ahuja that underscores the growing intersection of SI technology and human-centric design.