Pular para o conteúdo principalAWS Startups
Idioma do conteúdo
No momento, nem todo o conteúdo está traduzido.
  1. Aprenda
  2. Simile is creating digital twins for real-world results, with AWS

Simile is creating digital twins for real-world results, with AWS

Como estava esse conteúdo?

Simile develops AI-powered models of human behavior that simulate how individuals and populations might respond in different real-world scenarios. Built entirely on AWS, the startup uses Amazon Elastic Container Service (ECS), Amazon Simple Storage Service (Amazon S3), and Amazon Aurora to power its data-intensive platform, while AWS Activate Credits have helped it train and host increasingly sophisticated models, enabling organizations to make more informed decisions.

Artificial intelligence has unleashed a world of creators. Using AI tools has made it easier to build almost anything you can imagine. Now, the bottleneck has shifted upstream. The question is no longer whether you can you build something, but what you should build, for whom, and why. Simile helps organizations answer those questions by simulating how people will respond before decisions are made. 

Simulating the real world

Simile's vision is “quite straightforward,” explains Co-Founder and CEO Joon Sung Park: “to build a simulation engine that captures the complexity of human behavior.” This ambition can be traced back to Park’s experience as a researcher at Stanford. In 2023, his team created a simulated town populated by AI agents that would “wake up in the morning, do their routines, go to work, have relationships,” and essentially behave like real people in the real world, he explains.

The idea gained momentum when “our corporate partners who visited Stanford saw this demo and saw the opportunity to run all their studies in simulation,” he recalls. Sensing an opportunity, Park co-founded Simile alongside Lainie Yallen, Percy Liang, and Michael Bernstein. Working with AWS, the company created an enterprise-ready solution and quickly gained momentum.

Within months of launching, Simile grew its revenue fivefold and had run tens of millions of simulations for Fortune 100 companies. The startup describes itself as an AI lab, creating foundation models of human behavior that can simulate people's actions and attitudes “across any real human setting and across different experiments,” explains Park.

Today, its solution is used by some of the world’s largest companies “for everything from making better decisions around product assortment, to deciding what new market to enter, to really answering macro decisions that companies are trying to make around future strategy,” says Yallen.

Instead of running focus groups or real-world pilot products (which can be costly), or launching a new product without prior testing (which can be risky), organizations can take an idea, “whether that be text, video, image,” says Yallen, “and then test those against any end constituents of people,” in a simulated environment. In practice, this could be a financial services company wanting to predict the kinds of questions equity analysts will ask on an earnings call or a pharmaceutical firm needing to find out how a doctor might respond in reaction to a new drug.

The right data and the right training mechanism

To power its solution, Simile uses generative AI to create digital twins that simulate human behavior, preferences, and attitudes. This is a “unique” take on model building that takes a different approach from many organizations building AI. “In today's AI landscape you have companies that are basically creating machines that are amazing at rational thinking: super intelligent AI that can do mathematics, natural sciences, and coding,” says Park. Simile’s mission, in contrast, was to create models that can represent people's behaviors, attitudes, and tastes in all their human diversity. To do so required “a combination of the right data and the right training mechanism,” he says.

For this, Simile needed a partner “that could power the kind of complex applications and ecosystem that we have.” The startup deals with huge volumes of data: “Many of our customers are some of the largest corporations in the world with tens of millions of active users,” says Park. “They also want to leverage that data to give them the best insight into what their customers and users might need so they can best support their users.”

To achieve this, Simile combines data sources to build models of human behavior. The startup collects qualitative data, including interviews and surveys, to capture people’s preferences and perspectives; behavioral data from experiments and randomized control trials conducted in simulated environments; and data provided by its customers. Using this information, it creates digital twins that reflect the characteristics and behaviors of real people, which can then interact in simulated environments to test how individuals and populations respond.

Trying to gain similar insights and achieve similar results traditionally required companies to experiment with real participants, a process that can take weeks or months and makes it difficult to study large or hard-to-reach populations. Simile’s approach overcomes these constraints, as Yallen explains, “Today with simulation, you can test a new market entry into pretty much any available geographic market and actually see the potential downstream impacts of that, what cannibalization it might have against other products or brands and how that might impact your home portfolio, and then prioritize accordingly.”

Security and scalability

As Simile's business has grown, so has the complexity of the data behind its simulations. The platform is built entirely on AWS infrastructure, giving the startup the compute, storage, database, and security capabilities it needs to support enterprise workloads.

Simile uses Amazon ECS as its primary compute layer, Amazon S3 for storage, and Amazon Aurora as its primary database. Together, these services support the ingestion and storage of research data from more than 20,000 participants every week. Multi-region deployment helps Simile maintain high availability while meeting requirements around data residency and regulatory compliance.

Security is equally critical. Simile works with some of the world's largest enterprises, including organizations in highly regulated industries, so its platform has to meet strict requirements around customer and research data. AWS security controls, including VPC isolation, IAM-based access management, and encryption at rest and in transit, help Simile build these protections into its platform.

“AWS enables us to work with some of the largest companies in the world in a way that they know is secure and that affords us the kind of scale and pace required to go to market,” says Yallen.

As Simile develops increasingly sophisticated foundation models and generative AI agents, its infrastructure requirements continue to evolve. One challenge has been securing the GPU capacity required for increasingly intensive model training at the right time and cost. “This is where we've been very excited about the partnership that we've been forming with AWS,” says Park.

AWS supports Simile across its data and model lifecycles, from data collection and model training to hosting custom models for enterprise partners. “AWS can power that kind of complex application and ecosystem that we have across all of the data collection that we do. The model training, the hosting, the custom model training,” says Yallen.

Simile’s growth has also been supported by AWS Marketplace. “It’s been a great way for people to be able to reach us in the flow of their work,” says Yallen. The startup now counts some of the world’s largest organizations among its customers, including CVS Health, Wealthfront, Deloitte, and Gallup, as well as being accessible to smaller teams.

Indexing humanity

Simile partnered with AWS early in its startup journey. At this point, explains Yallen, “we hadn't raised a lot of capital. We hadn't really deployed the product and were in a stage where we needed to move quickly in a way that would suit our ambition.” This is a critical time for any startup, with progress often slowed by teams “spending too much time focusing on the thing that's not their job,” she says. Participating in AWS Activate allowed Simile to successfully navigate this potential hurdle, as Yallen explains: “The Activate program removed the distraction of administrative burden and funding required around the setup and enabled everyone to just focus on the core of what it is that they do.”

Designed for founders, AWS Activate provides eligible startups with up to US $200,000 in AWS Activate Credits to offset the cost of solutions and services, as well as access to exclusive offers, training, resources, support, and more.

Simile has already raised hundreds of millions of dollars and reached a multi-billion-dollar valuation, reflecting the growing demand for its AI-driven simulations. But this is just the beginning for the company. “With AWS, we are in the first innings of seeing simulation get deployed with companies around the world,” says Yallen. “In the next few years, this kind of technology powered by Simile is going to become a lot more ubiquitous.”

The company will continue working toward its ambitious vision of building a simulation engine that can model how people think, behave, and respond across virtually any real-world scenario. Achieving this will require Simile to expand its simulated populations across the US and international markets, collect and process huge volumes of data, and run simulations at massive scale. As Simile grows, AWS will continue to provide the scale, security, infrastructure, and guidance needed to support its ambition and help make simulation a tool for answering questions that were once impossible to test. “Ultimately, the most powerful technology is one that doesn't make your job easier,” says Park. “It's the one that allows you to do things that you could not imagine doing in the past.”

Como estava esse conteúdo?