“We’re Preparing Students for Real ML Roles”: Rajeev Rastogi on the Amazon ML Summer School

When Amazon's ML Summer School began in 2021 with 300 learners, the goal was to make the fundamental concepts of AI and ML accessible to would-be engineers. Fast-forward five years and nearly 10,000 learners, and the 2025 edition represents a new chapter in this expansion.

Amazon ML Summer School has integrated large language models, responsible AI, and problem-solving over simple theory - and done so intentionally with a push for diversity, inclusion, and geographic equity. Now that AI is transitioning from hype to implementation, Amazon's approach to pedagogy raises a broader question for the industry: how do you teach AI to not only code but also solve?

Rajeev Rastogi, VP of Machine Learning with Amazon, explains how the model is responding to industry needs, modernizing pedagogy, and enhancing/deepening and diversifying the local AI talent pipeline across India.

Five editions in—how is the Amazon ML Summer School 2025 different?

What does Amazon’s ML talent pipeline actually look like? 

Our approach towards machine learning talent pipeline is multifaceted. We have had to get creative and think long-term because the demand for ML talent continues to outpace supply globally, and particularly in high-growth markets like India. Our pipeline begins with broad educational initiatives like our ML Summer School and ML Challenge which reaches thousands of students each year. While these aren’t recruiting events, it's part of our commitment to democratizing AI education and expanding the overall talent ecosystem in India. We have seen that by investing in education broadly, we ultimately create a larger pool of qualified candidates, not just for Amazon but for the entire tech sector.

Our internship program has become a crucial pathway with specialized ML tracks that give students hands-on experience working on actual Amazon problems. What's unique about our ML internships is that interns aren't just working on simplified practice projects, they are contributing to real products and services that impact millions of customers. Last year, one of our interns developed an algorithm that significantly enhanced product discovery in specific Indian language searches, and we implemented it in production before their internship even ended.

We also invest heavily in internal mobility and upskilling. Some of our ML practitioners started in non-ML roles at Amazon and transitioned through programs like Machine Learning University and ML Gurukul, which provides Amazon employees with the skills to move into ML positions. This internal pipeline has become increasingly important as the field evolves.

Research represents another critical dimension of our pipeline. Our collaboration with academic institutions through funded research and faculty engagement helps us stay connected to cutting-edge developments while also building relationships with top talent. Many of our most innovative ML scientists joined Amazon after collaborating with us during their academic careers.

The challenges, of course, are significant. Competition for experienced ML talent remains intense, and the field moves so quickly that keeping skills current requires continuous learning. But our diversified approach has helped us build ML teams that combine deep technical expertise with business understanding and customer focus that make Amazon distinctive.

In what ways are Amazon’s internal ML systems shaping the pedagogy?

It's actually one of the most distinctive aspects of our broader educational initiatives including ML Summer School - the direct connection between what we are building at Amazon and what we are teaching. At Amazon, pedagogy is deeply shaped by how machine learning is applied at scale within our ecosystem. We’ve been pioneering in this space since 1998, starting with item-item collaborative filtering for recommendations—and today, ML touches nearly every aspect of our business, from product classification and logistics optimization to fraud detection and customer experience.  

This real-world integration informs us how we teach and train future ML professionals. Our pedagogy doesn’t isolate ML theory from application; instead, it is rooted in production-grade problem solving. Students and early-career talent are introduced to real-life use cases like AI-generated review highlights, Rufus—an expert shopping assistant that answers customer questions on shopping needs, products and comparisons, makes recommendations, and facilitates product discovery. We encourage learning that extends beyond algorithms to system-level thinking: how to scale models, deploy them responsibly, and align them with business goals. The rise of multimodal generative models—which combine text, images, audio, and video—has further expanded our educational focus to include human-centered design and interface innovation.

By exposing learners to these advanced, internal ML systems, we prepare them to build solutions that are not just technically sound, but impactful, scalable, and globally relevant.

Beyond the hype—what specific ML skills are truly in short supply in India today?

That's a question I think about constantly as we build our teams and educational initiatives.  First, contrary to popular belief, the gap isn't primarily in theoretical knowledge or basic ML techniques. India's educational institutions are producing graduates with foundations in mathematics, statistics, and traditional ML algorithms. What's genuinely scarce are practitioners with deep ML knowledge who can bridge the gap between theoretical understanding and business impact, people who can translate a vague business problem into a well-defined ML task, devise the appropriate approach applying and improving upon the state of the art, and implement a scalable solution that works reliably in production.

With over 34,000 women applicants till 2024, what has helped drive that kind of inclusive participation?

At Amazon, we believe that diversity drives innovation, and inclusive participation begins with creating equitable opportunities for all. When we launched ML Summer School, we wanted to ensure meaningful participation by all engineering students. Over the past five years, the program has grown into a highly sought after learning platform, with more than 1.3 lakh cumulative registrations, out of which 34,000 were women. This growing interest reflects the strong demand for accessible, high AI learning program in India.

Representation matters and we have women ML scientists from Amazon as instructors, mentors, and speakers - showing participants that there's a place for everyone in the field. Additionally, Amazon’s internal culture of inclusion sets the tone. When women see representation across all levels - including leadership - it reinforces the belief that they belong and can thrive. We empower our instructors on inclusive teaching practices and actively work to create a culture where no one feels they need to prove they belong. This is part of our long-term commitment to building inclusive tech talent pipelines.

What gives me the most hope is seeing the brilliant contributions of women participants in the program. Their unique perspectives and approaches to problem-solving strengthen the entire field of machine learning. When we remove barriers to their participation, everyone benefits—not just the participants themselves, but the entire tech ecosystem and ultimately the users of the technologies we build.

How do you design for scale without compromising depth?

When we started ML Summer School in 2021, we had about 300 participants and could provide an almost boutique learning experience. Now we are reaching thousands of students across India, and the tension between scale and depth is something we think about constantly. Technology has been crucial to our scaling strategy. We have invested heavily in our learning platform, which now provides personalized pathways based on each participant's background, progress, and performance. In each module of the summer school, we have included an interactive Q&A session with our instructors, who are ML experts, and provide in-depth answers to questions from participants.

We have also found that peer learning is a powerful multiplier. We have created structured opportunities for participants to learn from each other through discussion forums, peer code reviews, and collaborative projects. This not only scales our teaching capacity but often leads to deeper learning as students articulate concepts to their peers.

Our content development approach has evolved as well. Rather than trying to create all materials from scratch as we scale, we have adopted a modular approach where we develop high-value, Amazon-specific content for the areas where we have unique expertise and then curate the best existing resources for more foundational topics. This allows us to focus our content development efforts where they add the most distinctive value.

We have learned some valuable lessons along the way. Early on, we tried to maintain exactly the same experience for everyone as we scaled. Now we are more thoughtful about where depth matters most and where standardization is appropriate. We have also recognized that different participants have different needs—some value breadth of exposure while others seek deep expertise in a specific area. Our program now accommodates both.

Data has been crucial to this evolution. We meticulously track learning outcomes and participant feedback across different program elements, allowing us to identify where quality might be suffering as we scale and make targeted improvements. This continuous feedback loop helps us maintain depth where it matters most.

Looking ahead, we are exploring how AI can help us further balance scale and depth. The tension between scale and depth will always exist, but I have come to believe it's not a zero-sum game. With thoughtful design, technology enablement, and clear prioritization, it's possible to reach more learners while still providing the depth of experience. That's the balance we continue to refine with each edition of ML Summer School.

As LLMs and autonomous agents gain traction, is it time to rethink how we teach ML from the ground up?

The rise of large language models and AI is transforming how we should teach machine learning. While traditional ML education followed a linear path from basics to advanced topics, today's landscape demands a different approach. The ML Summer School program exemplifies this shift by inverting their learning sequence - students get a hands-on experience on applications of foundation models along with underlying theory and mathematics.

Students learn about engineering prompts for LLMs, observing their capabilities firsthand, along with learning about transformer architectures and attention mechanisms. The program helps participants learn more about building models from scratch along with data quality, preparation, and effective prompt engineering. Responsible AI is woven throughout the curriculum, emphasizing ethics, safety, bias, and transparency. As the ML Summer School has evolved, strong ML fundamentals remain essential - they're simply taught differently now, connecting theoretical principles directly to real-world applications using foundation models.

What do most candidates misunderstand about ML jobs at Amazon—and what stands out instantly in a great applicant?

Many candidates assume that ML roles at Amazon are purely research-driven or confined to building experimental models. In reality, what sets Amazon apart is the deep integration of ML into real-world systems at massive scale. ML scientists here are expected to not only develop advanced algorithms but also collaborate with product, engineering, and business teams to solve customer problems.

A common misconception is that academic excellence alone is sufficient. While a strong theoretical foundation is important, what truly stands out in great applicants is their ability to translate research into scalable solutions. We look for students who demonstrate practical problem-solving, ownership mindset, and the ability to innovate under constraints.

We also look for ML practitioners who can effectively explain complex models to non-technical stakeholders, translate business requirements into technical specifications, and build the cross-functional relationships needed for successful ML projects. This last mile of ML implementation often determines whether a technically sound model actually delivers business value.

At Amazon, we focus on building ML talent that not only understands deep learning architectures but can also scale models safely and responsibly. Future ML practitioners must be equipped to navigate both research and deployment, understand limitations of generative models, and design systems that align with human values also.