Mirror Particle bets a 'world model' can beat LLMs at predicting you

A two-year-old San Francisco startup called Mirror Particle is building what its co-founder calls a "world model" of human behavior — a foundation model trained from scratch to simulate why people do what they do — and it is about to compete in TechCrunch's Startup Battlefield, with a first venture round reportedly close to closing.

Mirror Particle enters a market that has already produced some of the largest early-stage rounds of the past year. Simile raised $200 million at a $2 billion valuation. Aaru raised $88 million at a $1 billion valuation. And in January, humans& announced a $480 million seed round at a $4.48 billion valuation and launched Persimmon, a product for modeling human behavior.

Against that backdrop, Mirror Particle is arguing that the dominant technical approach in the field is fundamentally flawed.

What's wrong with using LLMs to simulate people?

Today, most human behavior prediction relies on large language models that are prompted or fine-tuned to roleplay as a target demographic. Abhivyakti Ahuja, Mirror Particle's co-founder and CEO, thinks that approach cannot work.

"It's like bringing a super soaker to Niagara Falls," Ahuja said. "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."

Her objection is structural, not incremental. "LLMs are modeling written language, but humans are made of visual perception, spatial reasoning, social intelligence," she said. Relying on language models, she argues, produces insights grounded in what humans don't notice — which misses the point when the goal is predicting human behavior.

Instead of adapting an existing LLM, Mirror Particle is building its foundation model from the ground up. The company sells brands an AI engine that predicts consumer behavior and the reasons behind it.

Why 'the changing person' is the core of the model

The startup's central design choice is modeling people as systems that evolve over time rather than as fixed profiles.

"We don't want to capture the static person," 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."

Even stability carries information. If a segment isn't changing, she added, "that's also a signal."

To build these evolving models of demographic segments, Mirror Particle combines:

• its clients' customer data
• current events
• pop culture
• social media
• additional proprietary data sources

The engine tracks how motivations shift as a segment moves through experiences. Much of its focus sits on "revealed behavior" — what people actually do, rather than what they say in surveys.

The long-term ambition is to become what Ahuja calls the "general layer for anticipating human behavior," moving from population-level analysis down to individual-level insights. "We just need a better model of humans if we're going to work alongside AI and with each other," she said.

What can it actually do for a brand?

Like its rivals, Mirror Particle's initial go-to-market targets the departments that already budget for this kind of work: market research and brand and product strategy.

The scope goes beyond copywriting. Mirror might help a beauty brand not just write better ad copy for makeup aimed at Gen Z, but determine whether that demographic wants the product 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."

The engine also surfaces the "why" behind current or future behavior — the motivations, constraints and context that justify a recommendation — so brands can evaluate the reasoning, not just the output.

One early pilot illustrates the difference. A well-known pet food brand wanted to know which imagery to put on its packaging to boost sales: chicken, beef, or vegetables. Mirror's technology found the brand was asking the wrong question. The imagery didn't matter. The real problem was that the brand had become so recognizable it was perceived as mass market and cheap, and sales would plateau until the company addressed that perception.

Who is building it?

The company's approach to modeling human cognition traces back to Ahuja's academic background in neuroscience and computer science. Originally from India, she studied at the University of Toronto, where the work of AI pioneer Geoffrey Hinton on neural networks shaped her thinking.

After school, she joined Amazon Robotics, building robots that build other robots. There she met her co-founders, Will Song and Thomson Yen. Song spent much of their career building sales personalization engines. Yen focused on using deep learning to study how AI agents understand human behavior.

Ahuja describes the model's developmental trajectory in human terms: "The way we see our model evolving is like how a baby learns about the world," she said, noting that babies progress from vision to language to body awareness to social intelligence.

Mirror Particle has already raised an angel round and says its first venture round is near closing.

Why the stakes are high

The funding wave behind behavior-prediction startups signals that brands are willing to pay for simulated audiences as an alternative to traditional market research, which is slower and depends on self-reported data that often diverges from real behavior. Whether a from-scratch world model can outperform LLM-based rivals with vastly more capital is the open question — and one Startup Battlefield's VC judges, who decide this year's winner on the afternoon of Thursday, October 15 at Disrupt in downtown San Francisco, will watch play out alongside dozens of other vetted startups.