Lavu builds Marty AI on Amazon Bedrock to recover cash for restaurants

Lavu was formed after its founders saw local restaurants struggling with legacy POS systems. With the aim of providing something faster, easier, and more scalable, the company launched the first iPad-based restaurant POS system on the Apple App Store. It became one of the world’s largest mobile POS systems for restaurants.

The more integrated Lavu became in its customers’ businesses, the more it saw other problems it could solve. Restaurants were running systems for POS, payroll, payment processing, and other operational functions, but those systems didn’t speak to each other. Cash was leaking from the businesses, but by the time the owners noticed, the money had already been lost.

For Saleem S. Khatri, CEO at Lavu, the problem was also personal. A trusted employee at his father’s business had been deceiving the company for years without attracting attention from anybody there, at its bank, or at its external accounting firm. This shaped Khatri’s desire to help restaurant owners identify irregularities costing them money that they lacked the resources to detect.

The goal was to build the analyst Khatri wished his father had, using AI on AWS to bring enterprise-grade oversight to midmarket restaurants.

Lavu worked alongside AWS to launch Marty AI, an autonomous cash recovery solution that reviews restaurant data to identify where cash might be leaking and turns verified numbers into plain-language guidance for owners and operators. “The AWS team helped us go from first integration to a live customer-facing product in just 4 months and has stayed engaged through every scale step since,” says Khatri.

Central to the solution is Amazon Bedrock, which lets Lavu build generative AI applications and agents at production scale using over 100 foundation models from leading AI companies, including Anthropic’s state-of-the-art Claude model. Lavu uses Anthropic’s Claude in Amazon Bedrock to analyze complex financial reports, identify key trends, and generate clear, plain-language summaries for restaurant operators. For containerization, the architecture uses Amazon Elastic Kubernetes Service (Amazon EKS) to build, run, and scale production-ready applications.

Marty AI starts by gathering data from a restaurant’s POS, labor, inventory, reservation, and other systems. Python then verifies each figure against source files before the numbers pass through Amazon Bedrock for interpretation. Marty AI then writes a “Morning Receipt,” a daily briefing that shows the recoverable cash in a dollar figure with recommended actions. The briefing is sent by SMS by 6 a.m. the next morning, showing owners where to focus their attention at the beginning of each day. “Marty AI gives restaurant operators access to so many more inputs and so much more data so they can do what humans do best: take the context and make a decision,” says Kyle Bruce, senior vice president of operations at Lavu.

Establishing trust with restaurant owners is critical, and the split between the verification and analysis phases is key. Marty AI interprets only validated data, reducing the risk of misreading financial performance. “One wrong number and the operator stops reading,” says Khatri. “So, Python computes every dollar, and Anthropic’s Claude in Amazon Bedrock interprets only verified numbers.”

Since Marty AI launched, it has driven Lavu’s transformation from a traditional POS provider to an AI-powered analytics business. Because the solution is technology agnostic, multilocation restaurant groups can use it regardless of whether they run Lavu POS, which has opened new revenue streams and created a standalone analytics line of business. Lavu also plans to bring the same cash recovery capability to other industries, including construction, clinics, and agencies.

Marty AI is experiencing double-digit growth week over week on Amazon Bedrock. It’s had no customer-impacting outages since launch, and every Morning Receipt has been sent on time. In an industry operating on thin margins, Marty AI helps multiple-unit restaurant groups identify where they’re overspending. For one customer with two locations, it identified $200,000 in recoverable cash after analyzing just 3–4 months of data. Alongside the financial benefits, customers are also seeing broader gains in confidence and reassurance.

“You have to think about the benefit of the benefit,” says Khatri. “The benefit is the money. But the other piece is they’re able to trust their team and they’re able to verify. And that gives them peace of mind and comfort. That’s really what I think they’re after—peace of mind.”