San Francisco startup Mirror Particle is betting that large language models are fundamentally ill-equipped to predict human behavior, choosing instead to build a new foundation model from scratch.
The race to decode the human mind has moved into the realm of multibillion-dollar valuations. In the last year. companies like Simile. Aaru. and the $4.48 billion Humans& have captured massive capital by promising to predict consumer actions. Yet. in a San Francisco office. the two-year-old startup Mirror Particle is betting that the rest of the industry is fundamentally chasing a ghost.
Most current tools rely on large language models (LLMs) instructed to mimic specific demographics. To Abhivyakti Ahuja, the co-founder and CEO of Mirror Particle, that approach is a category error. “It’s like bringing a super soaker to Niagara Falls,” she says. “LLMs have been trained on hundreds of billions of data points. How much can you influence its behavior by fine-tuning with such a small amount of data?. It’s still stuck in the past.”.
Ahuja’s critique rests on a simple, stark reality: humans are not just text. She argues that LLMs model written language. while real human behavior is rooted in visual perception. spatial reasoning. and social intelligence. Mirror Particle is attempting to bypass this limitation by building a “world model” from scratch—a system designed to simulate the motivations behind human actions and how they shift over time. While competitors focus on capturing a static consumer. Mirror Particle tracks longitudinal data to see how individuals change. what triggers those changes. and to what degree.
The startup’s engine pulls from a proprietary mix of client customer data, pop culture, social media, and current events. The goal is to isolate “revealed behavior”—what people actually do—rather than relying on the unreliable answers found in self-reported surveys. In one pilot. a pet food brand asked the software whether chicken or beef imagery on packaging would drive more sales. The model returned a surprising answer: the imagery was irrelevant. It identified that the brand’s perceived status as a cheap. mass-market commodity was the real anchor on sales. rendering the packaging debate moot.
This obsession with the mechanics of the brain traces back to Ahuja’s academic roots in neuroscience and computer science. Originally from India. she later studied at the University of Toronto. where she was shaped by the work of AI pioneer Geoffrey Hinton. Alongside co-founders Will Song. who specialized in sales personalization. and Thomson Yen. who focused on deep learning for AI agents. the team moved through Amazon Robotics before launching Mirror Particle.
The technical architecture mimics developmental stages: starting with vision. moving through language and body awareness. and finally landing on social intelligence. It is a long-term play to become a general layer for anticipating human behavior. eventually shifting from broad population trends to specific. individual insights. As Ahuja puts it. “We just need a better model of humans if we’re going to work alongside AI and with each other.”.
Having already secured an angel round and nearing the close of its first venture funding. the company is preparing for a high-stakes test. Next week, Mirror Particle will take the stage at TechCrunch Disrupt 2026 in San Francisco to compete in the Startup Battlefield 200. From October 13 to 15. the startup will face a panel of venture capital judges. with a final winner to be crowned on the afternoon of October 15.
Mirror Particle AI human behavior prediction Abhivyakti Ahuja Startup Battlefield TechCrunch Disrupt 2026 foundation models