The Center for Data Innovation spoke with Surojit Chatterjee, founder and CEO of Ema, a company that provides AI-enabled virtual assistants that can automate various professional tasks, from customer service to data analytics. Chatterjee spoke about how his professional experience in the tech industry inspired him to create Ema, how one AI assistant helps pharmacists make pre-authorization decisions in seconds, and his vision for the future of work with AI and human employees.
Martin Makaryan: What is Ema, and what inspired you to found your company?
Surojit Chatterjee: Ema is a versatile, universal AI employee that can support various roles within an organization. She can assist everyone from customer support specialists to data engineers and help across industries like healthcare, human resources, and law. The inspiration to create this universal AI employee comes from both my personal background and from my decades of experience in the tech industry. I grew up in a remote village in India and I had never seen a computer before college. But in college, I studied computer science and the instant gratification of building software—you write code, and you can immediately see it work—fascinated me. I researched machine learning, focusing on areas like voice and handwriting recognition, which were still unsolved problems then. But after working at several major tech companies, I started Ema after noticing a recurring problem in all these companies: employees, even the most talented ones, often spend too much time on tedious tasks. I wanted to leverage AI to stop smart, talented people that enterprises hire from spending their time on tasks that AI can help automate. This led me to create Ema (short for enterprise machine assistant), an AI assistant to take on manual, repetitive, or otherwise mundane tasks so that people can focus on more creative and valuable work.
Makaryan: How does Ema work?
Chatterjee: Ema is a universal AI assistant functioning as a conversational operating system. It consists of two main layers or products. First, we have AI employees that you can hire and deploy to take on specific tasks like answering service desk tickets, helping with healthcare authorizations, or drafting complex documents. These AI employees are pre-built and ready to work using generative AI and the data and instructions that enterprises provide. For example, an AI employee could manage HR operations by answering questions about company policies or handling vacation requests. The second layer is what we call the “factory,” where you can create custom AI employees. It is like ordering a custom-built car—you can specify the exact role you need the AI to fill, and we will create it for you using specialized agents. For instance, one agent might be responsible for searching documents, another for summarizing them, and another for interacting with applications. These agents work together like a human team, with each one acting as an expert in its own field.
The core of the technology we have designed is the generative workflow engine, which integrates multiple AI models, including public models like GPT, with our private, domain-specific models. This allows us to optimize for both accuracy and cost, making Ema highly effective for enterprise use cases.
Makaryan: What sets Ema apart from other AI assistants?
Chatterjee: Ema is unique in how we bring these agents together seamlessly. Our customers do not need any coding skills to create a custom AI employee—they can do this through a natural-language conversation. Our AI agents are also highly specialized in whatever task enterprises need them to focus on, which makes them very accurate and efficient. Our goal was to design an assistant that can truly act like a real employee, and the ability to create custom personas also reinforces the role that Ema can play in organizations as not just a chatbot to answer questions, but an employee. For example, Ema helps pharmacists make pre-authorization decisions in seconds by checking patients’ medical records and analyzing long drug policy documents. She can also quote the relevant sources, which helps pharmacists verify the accuracy of her answers.
Makaryan: How can organizations leverage their internal data when using Ema?
Chatterjee: While public AI models like GPT have access to a lot of data, they are not trained using all the data that a specific organization may need. Many enterprises have sensitive information specific to their business or their field, so we have designed Ema work with internal enterprise data. Our customers can upload their data when creating custom personas using Ema to create more narrowly tailored assistants that meet their particular needs. We can also use an organization’s data to train new models to improve accuracy. These models are private and only accessible to that organization. Lastly, Ema remembers past interactions with human employees and continuously learns from them, improving its performance over time. For example, one of our clients, a large immigration law firm, saw Ema become as good as or even better than their best paralegals within just a few weeks of use.
Makaryan: What is your vision for Ema’s future?
Chatterjee: I believe AI employees and human employees will work hand in hand in the future. Managers will oversee teams that include both AI and human workers. AI employees will be able to handle repetitive, but also complex tasks, while humans focus on creativity and strategy. I also see a future where traditional software applications become less relevant. Instead of using multiple apps, people will simply ask AI assistants to complete tasks. My vision is for Ema to be be at the forefront of this positive disruption, transforming how enterprises operate.
