Finding Your TASTE - Career Flavors in AI/ML/Data Science

Over the past two months at Columbia, I’ve observed considerable confusion among my fellow graduate students—and frankly, among dozens of students I know across the US—about navigating careers in AI, machine learning, and data science. This post is my attempt to clarify the landscape, not as some success story or influencer, but as a fellow master’s student reflecting on what I’ve missed over the years and what I’ve observed closely in the AI ecosystem.

I’m writing this because I want to be open about what I believe and what I’m trying to achieve, and how this perspective might help others. Luckily, I’ve been able to stay focused these past few months and feel I’m making the transition I’ve been wishing for through this master’s program.

The Problem: Subtle Confusion and Fear

There exists a subtle confusion, even a subtle fear, about careers, jobs, and internships. Sometimes we end up having no idea what’s going on or how to even go about it. The typical advice from senior students is often just: “Keep applying for interviews.”

I don’t really like this approach. While we might need to do that eventually, I strongly believe we should understand the possible options and different paths ahead of us first. We should continue working on something meaningful rather than just mindlessly applying to jobs—that approach can drain your energy and motivation.

This blog is grounded not in all the successes I’ve seen over the years, but actually in something I missed over the years that I want to share, along with observations from closely watching developments in AI and its allied fields over the years.

Understanding the Industry Landscape

Before diving into specific roles, let’s understand the different types of companies and teams working on AI, machine learning, and data science. You might wonder: why do we always say “AI, ML, and data science”—aren’t they the same? The reason is that there exist three different types of teams that do these things differently, with some overlap but distinct focuses.

1. Frontier AI Research Companies

Companies like Anthropic, OpenAI, Google DeepMind, Meta FAIR, Amazon AGI, and Microsoft Research are investing heavily in building foundational models. They continuously experiment and scale their research on different architectures and training methodologies—focusing on training improvements, test-time scaling, mixture of experts, Mamba, canonical encoder-based architectures, and more.

These companies are the frontiers of modern AI, trying to match the qualities and standards of human intelligence. Their work represents the cutting edge of what we consider “AI.”

2. Search and Information Retrieval Companies

The next layer includes companies not primarily focused on developing human-level AI models, but rather on information retrieval and context management. Companies like Viaduct, Glean, Perplexity, and Gina AI work on search, context management, and building scalable AI-powered products.

Perplexity, for example, has developed AI-based search products like their browser. These companies support the AI ecosystem but aren’t building the foundational models themselves.

3. Infrastructure and Systems Companies

Companies like IBM focus heavily on infrastructure and systems—working on chip-level innovations, cloud services like Watson XAI, and the hardware that powers AI. Other examples include AWS, Databricks, and various cloud providers. These companies sell infrastructure services, cloud computing, and the underlying systems that make AI possible.

4. Machine Learning-Heavy Companies

Here’s where I want to differentiate between companies focusing on “AI” versus “machine learning.” Many teams at Microsoft Research, Google Research, Uber, Spotify, Netflix, and others aren’t working on human-level AI, but they’re doing extensive machine learning work.

For example, if we’re trying to forecast something—like modeling a cyclone’s path or predicting financial markets—the model isn’t going to have what we currently define as “intelligence.” But it’s excellent at understanding patterns and uncertainty from historical data to help us predict the unknown.

Companies use machine learning extensively for:

• Trading and financial applications
• Fraud detection and cybersecurity
• Recommendation systems (Spotify, Netflix, Uber)
• Demand forecasting
• Personalization and user experience
• ETA predictions

Even Adobe has been using machine learning for years across its products. When I worked at Informatica and Captain Fresh, we built ML models for specific business use cases—like understanding growth stages of shrimp using deep learning models to handle noise-to-signal ratios.

5. Companies Adopting AI Through APIs

An interesting shift has occurred: many companies have stopped their own machine learning efforts and started building AI-based solutions by procuring services from frontier AI companies. At OpenAI Dev Day, I saw that Notion won an award for utilizing a trillion OpenAI tokens—meaning Notion is building features that depend heavily on OpenAI’s APIs.

Previously, these companies might have used their own ML teams for auto-completion, classification, or search. Now many potential use cases are developed by adapting models from frontier AI labs. However, these companies still maintain their own ML teams for use cases that can’t be easily solved with external AI APIs—like fraud detection or specialized classification tasks that require training on proprietary data.

6. Enterprise Companies with In-House Labs

Fortune 500 companies like Walmart (Walmart Labs), Home Depot, Amazon, and financial institutions do everything for their own businesses. They typically have three types of work:

• AI Engineering: Utilizing AI APIs, RAG (Retrieval-Augmented Generation), and agentic workflows
• Machine Learning: Demand forecasting, personalization, recommendation systems using transformer-based models and other architectures
• Data Science: From dashboards and reporting to sophisticated econometric modeling

These companies have ML problems in both front-end (customer-facing) and back-end (business operations) areas. Front-end includes things customers interact with—recommendations, search, ETA. Back-end includes forecasting, inventory optimization, pricing, and fraud detection.

7. Well-Funded Startups and Y Combinator Companies

Startups vary widely. Some like Thinking Machine and Luma Labs focus heavily on AI—training models, providing infrastructure, or building reasoning and image generation models. Others like Cursor aren’t building AI or doing extensive ML, but they’re expertly infusing AI into their products, which requires scaling such experiences to many users through excellent software engineering.

Notion is similar—it may not require many ML engineers but needs many software engineers to build and scale AI-enabled features. And some startups might not use AI or ML at all but succeed through excellent backend engineering, design, and product development.

Understanding Different Roles

Now let’s talk about roles. It’s crucial to understand what you want to do rather than what title you want to get.

Research Engineers

Research engineers bridge the gap between research and engineering. They take new ideas—which might be very theoretical but proven through initial prototyping or theoretical validation—and scale them up. They work on how to accelerate research experiments and test them at larger scales.

Research engineering roles exist not just at frontier labs but also at companies like Uber, Spotify, Waymo, and other tech companies working on cutting-edge ML applications for their specific domains.

Machine Learning Engineers

The role of ML engineers varies significantly by company:

At business-focused companies (Uber, Spotify, Netflix, Waymo): ML engineers develop, train, and scale models for business-specific use cases like recommendations, ETA predictions, and personalization. These problems are highly complex and require contextual, higher-order solution design. You can’t simply use XGBoost or out-of-the-box models—you need deep understanding and often custom solutions.

At frontier labs and top tech companies: ML engineers might focus more on deployment, infrastructure, and scaling trained models into production systems.

The key distinction: ML at companies like Spotify (for recommendation systems, ETA) is about mining patterns from the company’s own data, which requires deliberate training of models. This is quite different from the ML happening at AI companies, which focuses on matching or exceeding human-level intelligence.

Data Scientists

Data science is a broad spectrum with vastly different tasks depending on the company:

• Some data scientists just create dashboards, crunch numbers, make reports, or conduct hypothesis testing and A/B testing
• Other data scientists do everything—they might have the title “data scientist” but work on full ML pipelines
• Economist-oriented data scientists (common at Spotify, Uber, Suno.com) focus heavily on behavioral research, causal inference, and understanding how people make decisions in marketplaces

For example, data science teams led by economists study how songs get discovered, how marketplace dynamics work, and what nudges can improve engagement. Without studying existing patterns, you can’t develop better strategies to make people consume more, engage more, or stay active on your platform.

Machine Learning Systems Engineers

Some companies hire specifically for performance optimization—people who work on GPUs, kernels, and low-level optimizations to make ML systems run efficiently.

Software Engineers (AI-Enabled Products)

These engineers may not know extensive ML theory but excel at building software that integrates AI capabilities. They work on products like Perplexity’s search, ChatGPT’s interface, agent frameworks, or infrastructure that provides sandboxes for AI applications. They build systems that enable and scale AI models into usable products and services.

The Full Spectrum of Work

Looking at this exhaustively, we can see the full spectrum from:

• Contributing to frontier AI research (new architectures, scaled models)
• Supporting frontier AI through infrastructure, scaling, and performance optimization
• Working on data infrastructure for model training
• Building systems for ML deployment and scaling
• Solving business problems with ML (recommendations, fraud detection, forecasting)
• Data science work ranging from analysis to sophisticated econometric modeling
• Software engineering to build and scale AI-enabled products

Finding Your Path

Now that you understand the landscape, here’s my advice:

First, identify what genuinely interests you:

• If you’re interested in econometrics and behavioral science: Focus on solid statistics, probabilistic modeling, causal inference, and experimental design. Pursue data science roles at companies with strong research cultures.

• If you want to solve business problems with ML: Focus on understanding how to translate business problems into ML problems. Study recommendation systems, fraud detection, forecasting, and learn deep learning for various use cases—not just LLMs, but the broader landscape.

• If you’re passionate about frontier AI and generative models: Go all-in on LLMs, diffusion models, and understand pre-training, mid-training, and post-training. As students, we have particular opportunities in test-time compute, evaluations, and empirical research on how reasoning works and where it fails.

• If you’re drawn to systems and infrastructure: Focus on distributed systems, GPU programming, performance optimization, and the engineering side of scaling ML.

Second, explore before you commit:

Don’t remain clueless or maintain only an abstract understanding of ML and data science. Look at what different companies are doing through their engineering blogs (check out Uber, Pinterest, Netflix, Spotify engineering blogs). Understand what specific teams work on.

Third, develop a strong objective:

Sticking with everything won’t work. You need to focus on a specific role or specific horizon of problem statements. Build depth in your chosen area.

My Own Journey

You might wonder what I’m trying to become. Having worked in both data science and machine learning engineering roles—on LLMs and beyond—I’m now focusing more on research engineering. I’m keeping a strong objective and maintaining focus toward this specific role and problem horizon.

Final Thoughts

The landscape of AI, ML, and data science careers is vast and nuanced. The key is understanding that:

1. Titles can be misleading - focus on what you’ll actually be doing
2. Companies differ significantly in how they approach AI/ML work
3. There are multiple valid paths - from research to engineering to applied work
4. Specialization matters - trying to do everything will dilute your efforts

Rather than just blindly applying to jobs, invest time in understanding these distinctions, exploring what resonates with you, and building genuine expertise in your chosen direction. Work on meaningful projects, stay focused, and let your work speak for itself.

I hope this helps provide some clarity. Let’s work towards our goals with clear objectives and focused effort.

---

This is a reflection from a fellow master’s student navigating the same journey. I’m learning as I go, and I hope these observations help you find your path too.