Today, we have the privilege of speaking with Nitin Gupta, an Engineering Leader at Meta, whose groundbreaking work in AIOps has had a lasting impact on Meta and across the industry. Following his talk at the prestigious System@Scale conference and his thought-provoking blog post, "The Evolution of AIOps at Meta: Beyond the Buzz," we sat down with him for an interview earlier this year.
Nitin is a veteran leader at Meta and has worked there for over 8 years. He has been working in the AIOps space at Meta for almost five years, and we have only been able to get the full story of his work now. His work has not only transformed how Meta approaches system reliability but also sparked innovation in broader areas like Machine Learning (ML) Debugging and rethinking Service-Level Objectives (SLOs), where others in the field have built upon his breakthroughs.
Nitin, could you tell us about your background and how you became involved with AIOps at Meta?
Nitin: I started working at Meta (fka Facebook) in 2015. Before that, I worked at AWS for 3.5 years. In both companies, I witnessed the complexity of large distributed systems. My journey in AI/ML began in 2019 when I was tasked to lead the AIOps initiative at Meta to simplify the investigations of incidents at Meta. I bootstrapped this initiative with a small team to do a quick prototype, and later, it expanded to a full-fledged business unit at Meta.
This is when I dove into AIOps—combining AI to tackle the massive volume of operational data, incidents, and anomalies in a way that engineers alone couldn't manage efficiently. Applying these technologies for root cause analysis was an innovative and novel approach, especially at the scale of Meta, where we have 100s of billions of time series, Petabytes (PBs) of logs, and thousands of events every minute to analyze within seconds to identify the root cause of an incident.
At System@Scale last year, I shared how our approach to AIOps is shaping the future, ensuring that we stay ahead of operational challenges as systems grow more complex.
In your talk, you highlighted the unique challenges at Meta for AIOps. Could you explain how your approach has transformed Meta's operations in a way that truly delivers value?
Nitin: Meta's Observability Systems had evolved over the years with data fragmented across multiple planet-scale systems, which made it extremely difficult for engineers to investigate issues. In a world with increasing complexity of systems and ever-growing telemetry data from over 2 billion users worldwide, engineers were left looking for a needle in a haystack.
AIOps, for us, has been about cutting through this noise. Rather than just monitoring for issues, we've created systems that actively learn from data patterns, predict failures, and automate the resolution process. It's this proactive approach that has significantly reduced downtime and improved system reliability. The real value comes from the time it saves engineers, freeing them to focus on more strategic challenges rather than firefighting.
You've demonstrated extraordinary ability, not only in developing these systems but also in leading teams to scale this work. Can you speak to the broader impact your work has had on the industry?
Nitin: It's been exciting to see how our efforts at Meta have rippled outward. Teams across the company and industry have taken the foundational work we did in AIOps and applied it to other critical areas. For example, ML Debugging has become much more streamlined, thanks to AIOps techniques that allow for real-time analysis of data from ML models in production. Similarly, my work has simplified existing workflows at Meta, like automating SLO attribution, which has empowered teams to ensure services meet performance targets without constant manual intervention. The team at Meta that built the ML Debugging system called Hawkeye based on my work recently published their findings on Meta's engineering blog. I encourage your readers to check it out. It's incredibly gratifying to see these concepts scale and solve bigger, more complex problems.
Why do you think your work has had such a broad influence, particularly in areas like ML Debugging?
Nitin: A lot of it comes down to solving problems that are universally challenging. Whether you're managing an AI model, optimizing for system performance, or ensuring service reliability, these are all areas where complexity grows exponentially. We focused on addressing fundamental data problems and ensured from the start that our systems were built for scalability and future expansion. Our team published research papers on these foundational technologies, like Fast Dimensional Analysis of Data.
Looking forward, what excites you about the future of AIOps and its potential applications?
Nitin: The future of AIOps is incredibly exciting because we're only scratching the surface of what's possible. I see a world where systems are not only self-healing but also self-optimizing—where AIOps enable predictive interventions that maintain system health before any human even notices a problem. There's also a huge opportunity in applying these principles beyond traditional Technology infrastructure into other fields where operational complexity needs to be tamed. Large Language Models (LLMs) and Foundational Models (FM) are only making this space more exciting.
What advice do you have for industry leaders and companies who are starting out in the field of AIOPs?
Nitin: Focus on the real operational pain points, not just what's trendy. AIOps isn't about adopting AI for the sake of it; it's about identifying and addressing the core issues your teams face, like system reliability and incident response. Build with purpose, and your initiatives will deliver value with AIOps.
Closing Thoughts
Nitin's pioneering work at Meta has redefined the possibilities of AIOps, making intelligent, scalable operations the standard rather than the exception at Meta. His innovations continue to shape the future of reliability, setting the stage for further breakthroughs in both AI and operational excellence.
