Meet Milind Naphade, Capital One's SVP of AI Foundations

We are thrilled to catch up with Dr. Milind Naphade, SVP, Technology, and Head of Capital One’s AI Foundations organization. Milind joined the company in May 2023 after a 30-year career in AI and tech, with stints at industry heavyweights IBM, NVIDIA and Cisco, bringing a wealth of knowledge of developing award-winning AI platforms and solutions that will be invaluable in powering innovative AI solutions at Capital One.

Since day one, Milind has been busy building out the new AI Foundations organization, creating partnerships with businesses and tech groups across the company, and contributing to Capital One's overall AI strategy. On the heels of his one-year anniversary in this role, we spoke with Milind about what attracted him to Capital One, his vision for AI, and how he sees AI transforming financial services and customer experiences in the years ahead. You can also learn more about Milind and his background on this recent Latent Spaces podcast.

At IBM Research, it was the spirit of innovation that matters to customers that truly stands out for me. This came through in almost every initiative, but the most memorable was our work on a Mobility Analytics platform and solution called Insights in Motion. This platform analyzed movement of people based on billions of anonymized telco data records in urban environments and applied these insights to solving various problems across multiple verticals. 

This entire initiative was born out of a chance conversation I had at the White House with an Assistant City Manager from a midwestern city. She had asked me whether the smart folks at IBM Research could look at mobile data and help her revitalize her city’s bus transit system. From that initial conversation, we created a solution platform that led to the cities of Dubuque, Iowa and Istanbul, Turkey winning awards for smarter transit implementations using the platform. The solution platform ended up assisting cities, retailers, telecommunications organizations and more with everything from complex transit system optimization to location-based marketing—all while paying respect to important guardrails and data privacy. That experience reminded me that a great idea can truly come from anywhere, and born from any conversation.

At Cisco, the commitment to customer-centricity was paramount. Here, we embarked on several initiatives, but what I'm especially proud of was a true force multiplier: augmenting the technical services team's ability to resolve customer issues. We used embeddings and our own homegrown implementation of a vector database to discover connections between cases to better solve them.

NVIDIA's "speed of light" ethos—putting aside benchmarks and being the best that you can possibly be—was a huge influence on my career, as was its culture of intellectual candor in acknowledging both successes and failures. The company's Metropolis program epitomized this, leveraging sensors to monitor in real-time and optimize all physical environments such as factories, buildings, retail stores and roadways in traffic networks, for example. That initiative underscored a core belief: impactful AI platforms must be built with humans as their focus. That mindset can foster thoughtful innovation at massive scale.

Every few years, I've made it a point to step out of my comfort zone and take on new challenges, whether it's within the same company or in a new environment. This has always remained within the broader canvas of AI and, in all of these contexts, I’ve focused on creating AI and ML platforms and solutions that have scientific and business impact. 

Within months of OpenAI releasing ChatGPT in November 2022, it was clear to me that the frontline of innovation was going to quickly shift from the underlying platform to solving business problems with it. Based on my past experience in other verticals, the only way for me to be part of that was if I was close to the business data and context needed to solve those problems. I found that the financial services sector was a natural next step for me, but it's a sector that is notoriously closed-off, especially when it comes to sharing data. If you want to innovate in financial services, you really need to be on the inside. 

That's why, when Capital One leadership approached me, and shared their vision, culture and aspirations, it became clear that this was a perfect fit. They understood the importance of tackling big problems and delivering customer value with technology and AI, had cultivated a culture that deeply values expertise, and were committed to innovation over the long term. They wanted me to come in and leverage all my expertise and experience to create a completely new organization that would be central to inventing the AI technology that allows all the Capital One lines of business to leverage AI and have a meaningful impact for millions of customers as well as thousands of associates.

I began with a deep dive into understanding the intricacies of the business, its challenges and the numerous opportunities where artificial intelligence can make a transformative difference. This understanding is crucial as it informs the design of our experiments, the AI platforms and the strategies we employ to tackle our prioritized use cases.

A significant part of my focus has been on building the AI Foundations organization—a task that is as strategic as it is reflective of our larger mission. Alongside the rest of our technology organization at Capital One, we’re continuing to build a diverse and world-class AI team (Capital One was recently named the #1 bank for AI talent by Evident Insights), all while architecting a work environment that fosters innovation, thoughtfulness and critical thinking at the forefront of some of our industry’s most interesting challenges.

A big part of this has been launching our new Applied Research job family to help Capital One continue to accelerate the adoption of state-of-the-art AI research into our business to enhance the customer experience and drive transformational business outcomes. Some of the key research areas we’re focused on range from anomaly detection and behavior models to large language models (LLMs) and graph networks. We also recently launched a related Applied Research PhD internship program for AI research, and are excited to welcome our first cohort of interns this summer.

Additionally, I’ve been heavily involved in the launch of our newest working spaces in California’s South Bay (San Jose, specifically), which is an exciting addition to our presence in the Bay Area and another way for world-class tech and engineering talent to collaborate in-person on some of the most challenging AI problems in our industry.

We are making great progress across all of these areas: Building out our AI Foundations organization - which has added some incredible talent over the last several months - as well as beginning to develop foundational AI capabilities to support critical use cases designed to improve customer and associate experience.

My work across different industries—smart cities, transportation, telecommunications, retail—has given me a unique perspective on the universal patterns of customer behavior. Understanding customer interactions at various touchpoints is a common challenge across these sectors. For instance, knowing the right moment to present an offer or advertisement can significantly boost customer engagement and deliver a better experience for the user, a concept at the heart of location-based marketing strategies.

Similarly, being able to respond to customer needs effectively, whether it’s during a direct interaction or through anticipating their requirements, is crucial. While the high-level patterns of customer engagement remain consistent, the specifics - like the data granularity and event types - can vary greatly from one industry to another.

In the realm of financial services, the sheer volume of textual data—from policy details to customer communications—is staggering, but it also presents a rich tapestry of information. The transactional data adds another layer of depth, and it has temporal insights, too. The transformer neural network architectures are particularly adept at interpreting these complex data sequences. These models, foundational to advancements in vision, conversational AI, and speech technologies, are equally applicable and potent in the financial sector, providing us with sophisticated tools to interpret and act on a wide array of customer data.

The frontier that really captivates me in AI research is the infusion of reasoning capabilities into existing AI architectures. I am a big fan of Yann LeCun’s vision to realize Autonomous Machine Intelligence, where he proposes that humans have an inherent “world model.” His vision is to create machines that can learn internal models of how the world works so that they can learn much more quickly, plan how to accomplish complex tasks, and readily adapt to unfamiliar situations. I believe that bringing reasoning based on the world model will be essential to get to the next groundbreaking shift in AI. Applying this to the financial sector, where decision-making is paramount, is particularly exciting to me.

Smaller language models like Mistral/Mixtral as well as the Llama 3 series are also quite noteworthy. Despite their size, they're achieving results comparable to their larger counterparts, which opens up possibilities for more efficient and scalable AI solutions in financial services—especially where speed and accuracy are most important.

Multimodality in LLMs represents another intriguing development. LLMs that can interpret text, audio and visual content and swap outputs between media types are poised to level up customer interactions by providing much richer experiences.

Looking at the trends emerging from conferences like NeurIPS and NVIDIA GTC, it's clear that the research community is pushing towards more agile and contextually-aware AI. These advancements have the potential to redefine how we interact with and manage our financial health, making services more accessible and tailored to individual needs.

This summer marks three decades in which I’ve been working in the field of neural networks, artificial intelligence, machine learning and its applications to various domains. In the last three decades, I have been fortunate enough to work with and learn from some amazing people in some of the most cutting edge research institutions - The Center for Development of Advanced Computing, the University of Illinois at Urbana Champaign, IBM Research, Cisco and NVIDIA. 

Some of the pioneering and seminal work we have done in this context has led to a number of professional accomplishments, including earning multiple innovation and research awards for not only these organizations but also for the customers and partners who leveraged the solutions we developed (this includes for the cities of Dubuque and Townsville, Queensland, Australia).

I’m also proud of several personal accolades, including receiving the Param Award for my undergraduate thesis on neural networks and C-DAC’s distributed supercomputer as well as earning the IEEE CAS Outstanding Young Author award for a journal paper based on my Ph.D. thesis on factor graph framework applications for semantic video indexing. My research publications have also won additional best paper awards. 

All of these milestones are products of amazing teamwork and applying AI to solve various societal and business problems.  

Outside of work, I'm a huge cricket fan and take every possible opportunity to watch the Indian cricket team.

My other favorite pastime is to try to help people out—students, colleagues, friends—with their AI problems. Problem-solving and helping others is something that keeps me humble, gives me energy and helps me stay current on all things AI. 

If you want to join our team at Capital One to help solve some of our industry’s most pressing challenges, we’re hiring! Learn about our Applied Research roles and other open tech and AI roles here.