Surojit Chatterjee is the founder and CEO of Ema. Previously, he guided Coinbase through a successful 2021 IPO as its Chief Product Officer and scaled Google Mobile Ads and Google Shopping into multi billion dollar businesses as the VP and Head of Product. Surojit holds 40 US patents and has an MBA from MIT, MS in Computer Science from SUNY at Buffalo, and B. Tech from IIT Kharagpur.
Ema is a universal AI employee, seamlessly integrated into your organization’s existing IT infrastructure. She’s designed to enhance productivity, streamline processes, and empower your teams.
Can you elaborate on the vision behind Ema and what inspired you to create a universal AI employee?
The goal for Ema is clear and bold: “transform enterprises by building a universal AI employee.” This vision stems from our belief that AI can augment human capabilities rather than replace workers entirely. Our Universal AI Employee is designed to automate mundane, repetitive tasks, freeing up human employees to focus on more strategic and valuable work. We do this through Ema’s innovative agentic AI system, which can perform a wide range of complex tasks with a collection of AI agents (called Ema’s Personas), improving efficiency, and boosting productivity across countless organizations.
Both you and your co-founder have impressive backgrounds at leading tech companies. How has your past experience influenced the development and strategy of Ema?
Over the last two decades, I’ve worked at iconic companies like Google, Coinbase, Oracle and Flipkart. And at every place, I wondered “Why do we hire the smartest people and give them jobs that are so mundane?.” That's why we are building Ema.
Prior to co-founding Ema, I was the chief product officer of Coinbase and Flipkart and the global head of product for mobile ads at Google. These experiences deepened my technical knowledge across engineering, machine learning, and adtech. These roles allowed me to identify inefficiencies in the ways we work and how to solve complex business problems.
Ema’s co-founder and head of engineering, Souvik Sen, was previously the VP of engineering at Okta where he oversaw data, machine learning, and devices. Before that, he was at Google, where he was engineering lead for data and machine learning where he built one of the world’s largest ML systems, focused on privacy and safety – Google’s Trust Graph. His expertise, particularly, is a driving force to why Ema’s Agentic AI system is highly accurate and built to be enterprise ready in terms of security and privacy.
My cofounder Souvik and I thought what if you had a Michelin Star Chef in-house who could cook anything you asked for. You might be in the mood for French today, Italian tomorrow and Indian the day after. But irrespective of your mood or the cuisine you desire, that chef can recreate the dish of your dreams. That’s what Ema can do. It can take on the role of whatever you need in the enterprise with just a simple conversation.
Ema uses over 100 large language models and its own smaller models. How do you ensure seamless integration and optimal performance from these varied sources?
LLM’s, while powerful, fall short in enterprise settings due to their lack of specialized knowledge and context-specific training. These models are built on general data, leaving them ill-equipped to handle the nuanced, proprietary information that drives business operations. This limitation can lead to inaccurate outputs, potential data security risks, and an inability to provide domain-specific insights crucial for informed decision-making. Agentic AI systems like Ema address these shortcomings by offering a more tailored and dynamic approach. Unlike static LLMs, our agentic AI systems can:
- Adapt to enterprise-specific data and workflows
- Leverage multiple LLMs based on accuracy, cost, and performance requirements
- Maintain data privacy and security by operating within company infrastructure
- Provide explainable and verifiable outputs, crucial for business accountability
- Continuously update and learn from real-time enterprise data
- Execute complex, multi-step tasks autonomously
We ensure seamless integration from these varied sources by using Ema’s proprietary 2T+ parameter mixture of experts model: EmaFusionTM. EmaFusionTM combines 100+ public LLMs and many domain specific custom models to maximize accuracy at the lowest possible cost for wide variety of tasks in the enterprise, maximizing the return on investment. Plus, with this novel approach, Ema is future-proof; we are constantly adding new models to prevent overreliance on one technology stack, taking this risk away from our enterprise customers.
Can you explain how the Generative Workflow Engine works and what advantages it offers over traditional workflow automation tools?
We’ve developed tens of template Personas (or AI employees for specific roles). The personas can be configured and deployed quickly by business users – no coding knowledge required. At its core, Ema’s Personas are collections of proprietary AI agents that collaborate to perform complex workflows.
Our patent-pending Generative Workflow Engine™, a small transformer model, generates workflows and orchestration code, selecting the appropriate agents and design patterns. Ema leverages well-known agentic design patterns, such as reflection, planning, tool use, multi-agent collaboration, language agent tree search (LATS), structured output and multi-agent collaboration, and introduces many innovative patterns of its own. With over 200 pre-built connectors, Ema seamlessly integrates with internal data sources and can take actions across tools to perform effectively in various enterprise roles.
Ema is used in various domains from customer service to legal to insurance. Which industries do you see the highest potential for growth with Ema, and why?
We see potential across industries and functions as most enterprises have less than 30% automation in processes and use more than 200 software applications leading to data and action silos. McKinsey & Co. estimates that generative AI could add the equivalent of $2.6 trillion to $4.4 trillion annually in productivity gains (source).
These issues are exacerbated in regulated industries like healthcare, financial services, insurance where most of the last decades technical automations have not happened since the technology was not advanced enough for their processes. This is where we see the biggest opportunity for transformation and are seeing a lot of demand from customers in these industries to leverage Generative AI and technology like never before.
How does Ema address data protection and security concerns, especially when integrating multiple models and handling sensitive enterprise data?
A pressing concern for any company using agentic AI is the potential for AI agents to go rogue or leak private data. Ema is built with trust at its core, compliant with leading international standards such as SOC 2, ISO 27001, HIPAA, GDPR, NIST AI RMF, NIST CSF, NIST 800-171 To ensure enterprise data remains private, secure, and compliant, Ema has implemented the following security measures:
- Automatic redaction and safe de-identification of sensitive data, audit logs
- Real-time monitoring
- Encryption of all data at rest and in transit
- Explainability across all output results
To go the extra mile, Ema also checks for any copyright violations for document generation use cases, reducing customers’ chance of IP liabilities. Ema also never trains models on one customer’s data to benefit other customers.
Ema also offers flexible deployment options including on-premises deployment capabilities for multiple cloud systems, enabling enterprises to keep their data within their own trusted environments.
How easy is it for a new company to get started with Ema, and what does the typical onboarding process look like?
Ema is incredibly intuitive, so getting teams started on the platform is quite easy. Business users can set up Ema’s Persona(s) using pre-built templates in just minutes. They can fine tune Persona behavior with conversational instructions, use pre-built connectors to integrate with their apps and data sources, and optionally plug in any private custom models trained on their own data. Once set up, experts from the enterprise can train their Ema persona with just a few hours of feedback. Ema has been hired for multiple roles by enterprises such as Envoy Global, TrueLayer, Moneyview, and in each of these roles Ema is already performing at or above human performance.
Ema has attracted significant investment from high-profile backers. What do you believe has been the key to gaining such strong investor confidence?
We believe investors can see how Ema's platform enables enterprises to use Agentic AI effectively, streamlining operations for substantial cost reductions and unlocking new potential revenue streams. Additionally, Ema’s management team are experts in AI and have the required technical knowledge and skill sets. We also have a strong track record of enterprise-grade delivery, reliability, and compliance. Lastly, Ema’s products are differentiated from anything else on the market, it is pioneering the latest technical advancements in Agentic AI, making us the go-to choice for any enterprise wanting to add next-generation AI to their operations.
How do you see the role of AI in the workplace evolving over the next decade, and what role will Ema play in that transformation?
Ema’s mission is to transform enterprises and help every employee work faster with the help of simple-to-activate and accurate agents. Our universal AI employee has the potential to help enterprises execute tasks across customer support, employee support, sales enablement, compliance, revenue operations, and more. We’d like to transform the workplace by allowing teams to focus on the most strategic and highest-value projects instead of mundane, administrative tasks. As a pioneer of agentic AI, Ema is leading a new era of collaboration between human and AI employees, where innovation flourishes, and productivity skyrockets.
Thank you for the great interview, readers who wish to learn more should visit Ema.
#ADD, #Agent, #Agents, #Ai, #AIAGENTS, #AISystems, #Amp, #Applications, #Approach, #Apps, #Audit, #Automation, #Behavior, #Billion, #Building, #Business, #CEO, #Cloud, #Code, #Coding, #Collaborate, #Collaboration, #Collections, #Companies, #Compliance, #Computer, #ComputerScience, #Copyright, #CustomerService, #Data, #DataPrivacy, #DataPrivacyAndSecurity, #DataProtection, #DataSecurity, #Deployment, #Design, #Development, #Devices, #Domains, #Driving, #Easy, #Efficiency, #Ema, #Employees, #Engine, #Engineering, #Enterprise, #Enterprises, #Envoy, #Financial, #FinancialServices, #Focus, #Functions, #Future, #Gdpr, #Generative, #GenerativeAi, #Global, #Google, #Graph, #Growth, #Healthcare, #Hipaa, #Hired, #How, #HowTo, #Human, #Industries, #Infrastructure, #Innovation, #Insights, #Insurance, #Integration, #INterview, #Interviews, #Investment, #IP, #ISO, #ISO27001, #Issues, #It, #ITInfrastructure, #Jobs, #Language, #LanguageModels, #LargeLanguageModels, #Learn, #Learning, #Legal, #LESS, #Llm, #LLMs, #MachineLearning, #Management, #Mit, #Ml, #Mobile, #Model, #Models, #Nist, #Okta, #Onboarding, #One, #Oracle, #Orchestration, #Organization, #Organizations, #Other, #Parameter, #Patents, #Patterns, #Performance, #Planning, #Platform, #Play, #Prevent, #Privacy, #Process, #Productivity, #Profile, #RealTime, #Reflection, #Reliability, #REST, #Revenue, #Risk, #Risks, #Roles, #Safety, #Sales, #Science, #Search, #Security, #SecurityRisks, #Sensitive, #SOC, #SOC2, #Software, #SoftwareApplications, #Solve, #Stack, #Standards, #Strategy, #Streams, #Teams, #Tech, #Technology, #Template, #Templates, #Time, #Tool, #Tools, #Training, #Transform, #Transformation, #Transformer, #Tree, #Trust, #Vision, #Work, #Workflow, #Workflows, #Workplace
Published on The Digital Insider at https://is.gd/seMLGN.
Comments
Post a Comment
Comments are moderated.