Mirror Particle, a two-year-old startup based in San Francisco, is developing a foundation "world model" intended to simulate, predict and explain human behavior over time. The company argues that relying on large language models (LLMs) — typically prompted or fine-tuned to role-play target demographics — is an inadequate approach.
"It's like bringing a super soaker to Niagara Falls," said Abhivyakti Ahuja, Mirror Particle’s co-founder and CEO. Ahuja contends that LLMs are trained to model written language, while human decisions are strongly shaped by visual perception, spatial reasoning and social intelligence. Fine-tuning an LLM on a relatively small dataset, she warns, leaves the model "stuck in the past."
A different approach: a longitudinal world model
Rather than creating static consumer profiles, Mirror Particle is building a model that captures how people change over time, which triggers drive those changes and to what degree. Longitudinal data — including signals that someone is not changing — is treated as information.
The startup combines proprietary data sources: clients' customer data, current events, pop culture, social media and other contextual inputs. It models demographic segments as evolving systems and tracks how motivations shift as people move through experiences. Mirror Particle emphasizes "revealed behavior" — observing what people actually do — over self-reported survey answers.
Go-to-market and client use cases
Mirror Particle’s initial commercial focus targets areas where budgets already exist for behavioral insights, such as market research and brand or product strategy. Beyond helping a beauty brand craft ad copy that resonates with Gen Z, the company can help determine whether that demographic even wants the product in question.
"What if [the target demographic] doesn't want eyeshadow palettes?" Ahuja asked. "Maybe blush is a better option to go for if you want to sell a product to this market."
In an early pilot, a well-known pet food brand asked which imagery on packaging would boost sales — chicken, beef or vegetables. Mirror Particle’s analysis concluded the brand had posed the wrong question: imagery was not the core issue. The brand’s strong recognition had produced a mass-market, low-cost perception that would keep sales plateaued until that reputational problem was addressed.
Team background
Ahuja draws on neuroscience and computer science training; she studied at the University of Toronto and cites Geoffrey Hinton’s work on neural networks as an inspiration. After school she worked at Amazon Robotics, building robots that assemble other robots, where she met co-founders Will Song and Thomson Yen. Song has experience building sales personalization engines, while Yen focused on using deep learning to teach AI agents about human behavior.
Funding, competition and long-term vision
Mirror Particle has raised an angel round and says it is close to closing its first venture round. The startup will compete in TechCrunch’s Startup Battlefield 200 at TechCrunch Disrupt 2026 in San Francisco, which takes place October 13–15; the Startup Battlefield winner will be chosen by VC judges on the afternoon of Thursday, October 15, 2026.
The company's long-term ambition is to become a general layer for anticipating human behavior, moving from population-level analyses toward individual-level insights. "We just need a better model of humans if we're going to work alongside AI and with each other," Ahuja said.
Why this matters
Mirror Particle’s approach highlights a broader debate in AI research: whether language-derived patterns are sufficient to understand human decision-making, or whether multimodal, time-sensitive models that track visual, social and contextual change are necessary. The startup’s methods and early pilots suggest brands and product teams could make better decisions if they can see both observed behavior and the motivations behind it.
