How Genius AI Built a Growth Analyst for 125,000 Service Businesses on Mastra

Genius AI is a business-software company offering AI tools for local businesses. Its products, including GlossGenius and Reception, help automate bookings, marketing, customer communication, payments, staffing, and other administrative work for more than 125,000 businesses—salons, spas, and wellness studios—that generate $7 billion in sales through the platform each year.

Genius AI has built a Growth Analyst used by thousands of customers each month and is rolling out AI Receptionist and Marketer agents. Together, they help customers understand their performance, recover missed bookings, and identify measurable opportunities for growth.

From one-shot prompts to real agents

Danielle Cohen-Shohet's first business was freelance makeup artistry, run out of her Princeton University dorm room. A few years in, she understood the problems of service-based self-care businesses well: the owner is often doing the work, managing the calendar, answering customers, and trying to grow the business at the same time.

Danielle decided to start GlossGenius to solve these problems. Nearly a decade later, GlossGenius is the flagship product of Genius AI, bringing bookings, payments, marketing, and business management into one platform.

The next step was introducing AI features. "How can we do it in a way that is genuinely adding value? Not replacing anything, but enabling them to do things they couldn't before?" says Carlos Tamulonis, Staff Software Engineer, Genius AI.

Now, Genius AI has four different agents built on Mastra:

Growth Analyst: A data scientist and business consultant inside GlossGenius, connected to a customer's business data. It identifies opportunities to improve and compares performance with similar businesses in the same vertical or area.
Receptionist: A text and voice agent that helps respond and book appointments when the owner is unavailable. Mastra powers most of its capabilities.
Marketer: Helps with a business's presence across local and AI search and, over time, re-engages existing customers with more targeted outreach.
ACE, an internal SRE agent: Helps engineers get information about their production systems, day-to-day or during an incident, and act on it quickly.

Choosing Mastra

When Carlos joined Genius AI in early 2025 to lead the new AI engineering team, the company already had several LLM-powered features. They used one-shot prompts for narrow tasks, like turning a customer's availability, communicated in natural language, into structured data for an appointment waitlist. But there was no shared agent framework, tool-calling layer, or evaluation system.

As the team moved toward more dynamic agents, they wanted a framework that would fit their existing stack. Genius AI primarily builds with Kotlin and TypeScript, and didn't want to introduce a Python-only service that would separate AI development from the rest of product engineering.

Mastra's focus on TypeScript gave the team confidence that the JavaScript ecosystem wouldn't be secondary to Python. Mastra also had the breadth of features they needed in one framework: agents, tools, persistence, processors, guardrails, observability integrations, evals, voice support, and an open-source playground.

"Mastra was one unified surface. It integrated with Postgres, was easy to deploy, and we really liked the API. It was very straightforward." — Carlos Tamulonis, Staff Software Engineer, Genius AI

Making Mastra the heart of a shared AI service

Genius AI runs Mastra as a standalone service and wraps it with gRPC, since their internal services already use this protocol to communicate. Product traffic reaches the service through the API gateway, while Mastra handles the core agent runtime and its own PostgreSQL-backed persistence.

The team added a configuration layer that lets engineers define an agent in a TypeScript file. One object brings together the agent's model, prompts, structured-output schema, guardrails, processors, and observability settings.

Some agents are actual agents—they have tools and run an agent loop—and some are just one-shot prompts that still go through Mastra. As Carlos says, they're leveraging Mastra almost like an LLM gateway.

The AI system is connected to Genius AI's proprietary frontend system for streaming responses and rendering charts, dialog boxes, and other interactive elements. Product teams own the customer experience, and Mastra provides the shared AI foundation.

As the team expanded the analyst's scope, they introduced a specialized subagent that formulates SQL, runs the query, and checks data quality. The main agent can now answer a much broader range of questions without mixing all of that complexity into one prompt.

The platform also gained agent versioning and experimentation, customer-specific dynamic context, and explicit prompt-caching boundaries.

Treating evals as part of the product

"We realized early on that evals were what was going to make us a quality AI experience or just a random noise machine," says Carlos.

They started with a set of representative questions and structured checks: Did the agent select the right tool? Did it supply the right arguments? Did it interpret the result correctly? Over time, Genius AI added automated evals to CI so that changes to a prompt, model, or agent configuration could be checked for regressions before they reached customers.

The team reviews samples of real interactions, replays older traces against new prompts or models, and uses the results to guide iteration.

From separate agents to one business partner

The Growth Analyst has been in production since January 2026 and is used by thousands of users each month. The Marketer and Receptionist have begun rolling out to customers in earnest. The potential reach is much bigger with more than 125,000 businesses already on the platform.

Today, the three customer-facing agents are separate experiences. But the long-term vision is to make their capabilities feel like one business partner.

In the shorter term, Genius AI is exploring Mastra skills and progressive disclosure so that one agent can load the instructions and tools it needs for the task at hand. Sandboxes could give it a safe place to work through more complex jobs, while memory could help it carry useful context across sessions.

"Whenever I need to expand the agents a little bit, there's something in Mastra that can help me. Someone has already thought about it. I can turn it on instead of having to figure it all out myself." — Carlos Tamulonis, Staff Software Engineer, Genius AI