Journey

The Next Chapter

OpenAI represents the frontier of AI deployment at scale. ChatGPT serves millions of users with real-time inference requirements that push the boundaries of what is possible.

Why This Role

Every role I have had prepared me for this:

• Performance engineering taught me to optimize
• Netflix taught me scale
• PicPay taught me leadership
• Parasail taught me AI infrastructure

Now I am applying all of it to one of the most impactful AI products in the world.

What I am Learning

The scale of AI inference is unlike anything I have worked on. Real-time responses, massive model sizes, and user expectations that do not tolerate latency. It is the hardest performance problem I have faced.

Looking Forward

I believe AI will transform every industry. Being part of building that infrastructure—making AI faster, more reliable, more accessible—is the most important work I can do right now.

The Opportunity

Parasail sat at the intersection of my experience: platform engineering, infrastructure, and AI. The mission was to make AI deployment as simple as it should be—no contracts, no quotas, no cloud complexity.

The Challenge

Building infrastructure for AI workloads is fundamentally different from traditional applications. GPU compute is scarce, expensive, and requires different optimization approaches.

What I Built

Led the engineering organization building:

• GPU orchestration platform
• Inference optimization
• Developer tooling for AI deployment

Key Learnings

AI infrastructure is still infrastructure—the principles of reliability, observability, and developer experience apply. But the constraints are different, and understanding those differences is essential.

The Catalyst

The AI boom was not something I could ignore. I had spent my career building tools for engineers; AI was building tools that could help engineers build faster. I needed to understand it deeply.

What I Explored

• Local LLMs: Privacy-focused AI assistance without cloud dependencies
• Agent-driven development: Autonomous workflows where AI executes tasks
• Skill engineering: Teaching AI agents domain-specific capabilities
• OpenCode: Containerized development environments for AI-assisted coding

Projects

I started building practical projects:

• Custom skills for AI agents
• Containerized environments for local LLM development
• Tools connecting AI to development workflows

What I Learned

AI is not replacing engineers—it is changing how engineers work. The engineers who thrive will be those who can effectively collaborate with AI, not compete with it. This learning investment positioned me for my next role.

The Challenge

Moving from individual contributor to VP of a 200+ engineer organization was the biggest transition of my career. I had to shift from solving technical problems myself to enabling others to solve them.

Platform Engineering Strategy

I defined and executed a platform engineering strategy that:

• Reduced Change Lead Time by 20%
• Increased Deployment Frequency
• Improved developer satisfaction (CSAT) from 22% to 87%

Key Achievements

Developer Experience: Consolidated developer tooling into a unified Internal Developer Portal. Drove adoption through training and support.

Cost Efficiency: Established a FinOps practice across 16 Business Units, delivering over $30M in annualized savings.

Mobile Development: Spearheaded a Server-Driven UI framework that reduced time-to-market by 94%.

End User Experience: Focused on non-functional quality metrics, reducing ANRs by 88% and crashes by 80%.

What I Learned

Leadership is about multiplying impact through others. My job shifted from writing code to removing obstacles, making decisions, and communicating vision. The scale of impact was larger, but the work was less hands-on.

The Change

After nearly 9 years at Netflix, I was ready for a new challenge. Snowflake’s cloud data platform presented different performance problems—batch workloads, data pipelines, and query optimization.

What I Did

• Analyzed and optimized core service components
• Developed prototypes for system observability and performance analysis
• Collaborated on service deployment models to improve availability

Key Learnings

Different domains require different performance approaches. What worked for streaming did not always apply to data warehousing. This role expanded my toolkit and understanding of distributed data systems.

The Award

The Mullen Foundation was established in 1990 to honor the memory of J. William (Bill) Mullen, past president of CMG. The award recognizes individuals who demonstrate technical excellence combined with the ability to communicate effectively with each member of a wide audience.

The Presentation

“Real User Performance Monitoring at Netflix Scale” covered how Netflix usefully aggregates enormous quantities of high-dimensional data to understand and improve user performance for over 130 million users watching more than 140 million hours of content daily.

Recognition

“Martin is a dynamic Netflix speaker and this award is a testament not just to his technical excellence, but also to his ability to communicate. Martin’s presentation was judged outstanding by his audience and by our referee.” — Amanda Hendley, President and Managing Director of CMG

“Martin has been immersed in the study of user performance for the last few years. The results of his investigations are already having a measurable impact on our business.” — Ed Hunter, Engineering Leader at Netflix

About CMG

CMG (Computer Measurement Group) is a highly influential worldwide organization of IT professionals committed to digital transformation initiatives and best practices. Founded in 1974, it has built a highly skilled membership network over the past 40 years.

The Story

Época Negócios, one of Brazil’s leading business magazines, featured my career journey from a small town in southern Brazil to working at Netflix in Silicon Valley. The article highlighted how I became a Senior Performance Architect at Netflix, helping ensure millions of subscribers worldwide could stream their favorite shows without technical issues.

What They Covered

• My path from Nova Petrópolis to Silicon Valley
• Building a performance engineering practice at Netflix
• The unique challenges of optimizing systems at massive scale
• What it means to work on problems no one has solved before

The Article

De Nova Petrópolis (RS) para o Vale do Silício: conheça a história de Martin Spier, executivo da Netflix

The Problem

Traditional flame graphs show aggregated profiles, but performance issues often occur in specific time windows. A CPU spike lasting 200ms would be invisible in an hour-long profile. I needed a way to see patterns across time and zoom into specific ranges.

The Innovation

FlameScope combines two visualizations:

1. A subsecond-offset heat map showing patterns over time
2. Interactive flame graphs for selected time ranges

This lets you see when problems occur and what code is responsible, without filtering through irrelevant data.

Impact

FlameScope was presented at conferences and adopted by companies dealing with performance variance. It revealed patterns that were previously invisible to engineers.

What I Learned

The right visualization can make complex data intuitive. FlameScope taught me that tools should reveal patterns, not just display data.

The Need

Flame graphs are essential for understanding CPU profiles, but existing visualization tools were static SVGs. I wanted an interactive, web-native visualization that could be embedded in any tool.

What I Built

d3-flame-graph is a D3.js plugin that renders interactive flame graphs. You can zoom, search, and explore the call hierarchy. It handles large profiles gracefully and integrates easily into existing applications.

Adoption

The library was adopted by major projects:

• Apache Flink for job profiling
• Google’s pprof profiler
• Oracle’s Java Flight Recorder

Seeing my code used in tools I admired was incredibly rewarding.

What I Learned

Open source is about more than code. Documentation, examples, and responsive issue handling are what make a project successful. I also learned that a focused, well-documented library often has more impact than a larger, more complex tool.

The Problem

Engineers at Netflix needed high-resolution performance metrics, but existing tools were either too coarse-grained or required deep expertise to use. I wanted to make performance data accessible to every engineer, not just specialists.

What I Built

Vector was a web-based framework that connected to PCP (Performance Co-Pilot) on hosts and displayed real-time, high-resolution metrics. Engineers could see CPU, memory, disk, and network behavior at sub-second resolution without leaving their browser.

Open Source Impact

Vector was open-sourced in 2016 and adopted by companies worldwide. The project was eventually retired in 2021 as we migrated its capabilities to Grafana, but the core ideas—making performance data accessible and visual—lived on.

What I Learned

Building tools for other engineers taught me that usability matters as much as functionality. The best performance tool is the one people actually use.

The Opportunity

Netflix was in the middle of migrating to AWS and rebuilding its streaming platform. The scale was unprecedented—a single performance issue affected millions of users simultaneously. I joined to help build the performance engineering practice from the ground up.

What I Built

• Performance analysis tools and frameworks for microservices architectures
• Visualization tools for understanding system behavior at scale
• Practices and patterns that other engineers could apply

Impact

My work at Netflix shaped how I think about performance—not as an afterthought, but as a fundamental property of well-designed systems. I learned that at scale, you can’t fix performance problems after the fact; you have to build performance in.

The Challenge

Moving from Brazil to the US was a personal and professional leap. I had to adapt to a new culture, new ways of working, and systems operating at a different scale.

What Changed

At Expedia, I worked on performance for high-traffic travel booking systems. The scale was different from anything I’d experienced, and the stakes were higher—every second of latency meant lost revenue.

Key Learnings

• Performance at scale requires thinking about distributed systems differently
• Collaboration across teams is essential for end-to-end performance
• Building relationships matters as much as building solutions

This transition prepared me for the scale I’d encounter at Netflix.

The Beginning

Fresh out of university with a B.Sc. in Computer Science, I joined Dell’s IT division in Brazil. Performance engineering was a niche field, and I was drawn to the detective work of finding bottlenecks and making systems faster.

What I Learned

• How to profile Java applications and interpret the results
• The difference between perceived performance and measured performance
• That “slow” is rarely about one thing—it’s usually death by a thousand cuts
• How to communicate technical findings to non-technical stakeholders

This role gave me the foundation I’d build on for the next 15+ years.