Tips on how to build an AI product team: Q&A with Rukmini Iyer

 Rukmini Iyer Headshot

Rukmini Iyer is a local Bay Area leader and corporate vice president leading a fast-paced, innovative, sciences and infrastructure group dedicated to enhancing our Bing search platform.

Her team is a mix of machine learning (ML) experts, game theorists, economists, scientists, and software engineers that bring together unique skill sets to solve complex product problems. They are in charge of algorithmic components of the search advertising stack — from core Natural Language Processing efforts for query and ad understanding to machine learning of user and advertiser signals and incorporation in ad auctions.

We connected with Iyer to learn how her team functions and understand her approach to building a successful artificial intelligence (AI) product:

What is the focus of an AI product team?

One of the biggest misconceptions about working on an AI product is that you will be boxed into working on production models and code. No more conferences, state-of-the-art research, or large-scale models. This couldn’t be further from the truth.

Our team is a balanced mix of product and research. Because AI integrated within ads is so advanced, it needs to stay on innovation’s cutting edge. That is where the research and collaborations with Microsoft Research come in. At the same time, we deploy very large-scale models and ML data pipelines that must work within production latency and CPU constraints, this is also a significant software engineering challenge. We run experiments both offline and online to stress test these models before releasing them into the product.

There is a balance of pragmatism and a desire to have the best-in-class AI in the product. Most people don’t realize how fast that translation from research to product can happen, with the right team and infrastructure, it can be just a couple of months.

Do you look for specific areas of expertise when building a team?

AI is different from other engineering fields because it relies very heavily on math and computer science. Breakthroughs happen when you have a diverse team that learns and grows with the product. Computer science experts, statisticians, physicists, and strong math backgrounds are valuable additions to any team. We have people of all backgrounds working together, and it makes for a fun learning environment.

Things get old fast in the world of AI! You need people around you who love to learn.

What are some keys to success in creating a successful AI product?

You need to fundamentally rethink your business if you plan to rely on AI for doing the heavy lifting in your product. You must design your problem statement and comprehensive metrics and guardrails, while also setting up experimentation processes, robust data pipelines, and analytics.

It is also essential to empower your employees with the infrastructure and tools they need to succeed. You can’t build state of the art AI on sub-par infrastructure. Team leaders and managers need to commit to the whole process, not just the first step of building and deploying a model.

What’s next in AI that you’re keeping tabs on?

Causality, explain-ability, and having a responsible AI framework are all critical upcoming concepts in AI.

We’ve been training models from human-generated data. Now, the data we are using is created through AI and human interaction, optimized for product and business success metrics. This information has a different look and feel to it.

We are now asking ourselves, what does it mean to have AI learn from itself in real-world scenarios?

________

If you’re interested in joining Rukmini’s team and working on our AI products, opportunities can be found here: 

Sunnyvale specific openings: https://careers.microsoft.com/us/en/job/950341/Senior-Applied-Scientist 

WebXT organization open in multiple locations: