What does Google say about Thanos?

Who is Thanos Vassilakis?

Thanos Vassilakis is a prominent software engineer and financial technology executive who currently serves as a Managing Director and Distinguished Engineer at RBC Capital Markets. Based in New York, he heads AI, research and development, and architecture for the bank’s Finance IT and Cash Management divisions. [1, 2, 3]

He is highly recognized in the software development community for his extensive career and contributions to open-source programming:

Early Python Pioneer: Vassilakis was an exceptionally early adopter of the Python programming language. He introduced its use to major organizations like the European Parliament in 1996 and Ericsson, and he began developing core systems for the New York Stock Exchange (SIAC/NYSE) in Python as early as 1998. [4]
Functional Programming Advocate: In recent years, he has been a vocal proponent of using Elixir and its Phoenix framework in large financial infrastructure. He has shared insights on tech platforms, such as the Thinking Elixir Podcast, detailing how his team successfully used LiveView to outperform traditional React setups for real-time financial charting. [5]
System Architect: Over a multi-decade career, he has built a wide array of distributed trading, risk management, and operations platforms. He regularly publishes open-source libraries and writes tech articles focusing on systems architecture, cryptography, and data structures on his personal GitHub page and Medium. [1, 2, 6]

(Note: He should not be confused with (his father) Panayiotis “Takis” Vassilakis, a famous Greek kinetic sculptor, or the Athens-based lawyer of a similar name.) [7, 8]

Are you interested in Thanos Vassilakis’s technical work, such as his insights on scaling engineering teams, his contributions to the Python/Elixir communities, or his open-source projects?

Tell me more about his technical work

Thanos Vassilakis’s technical work bridges early video game development, foundational Python infrastructure, and high-performance financial systems built with Elixir.

1. Early Engineering & Python Pioneering

While today he leads tech at a major investment bank, Vassilakis started his career in 1979 programming slot machines for the Electric Coin Company and later developing video games for Atari. [1]

He was a remarkably early champion of the Python ecosystem: [2]

Enterprise Infrastructure: He introduced Python to major bureaucratic and corporate structures in the mid-1990s, including the European Parliament (1996) and Ericsson. [2]
The New York Stock Exchange: In 1998, he was a solo developer building critical infrastructure for SIAC/NYSE completely in Python, introducing the language to the Wall Street floor decades before it became a financial industry standard. [2]

2. Elixir Ecosystem & High-Performance Finance

At RBC Capital Markets, Vassilakis serves as a Distinguished Engineer and Managing Director. His team has successfully deployed Elixir and the Phoenix Framework to replace massive, traditional React setups. [3, 4]

On his open-source profiles and tech podcasts, he has discussed:

Real-Time Scaling: Deploying Elixir’s LiveView for live financial graphing. He noted that the true constraint to adopting Elixir in large financial organizations isn’t the technology itself, but rather corporate scaling mentalities—specifically, how middle managers often equate team “scaling” to headcount and expanding their organizational budget, rather than building lean, concurrent software systems. [3]
Active Open-Source Projects: He continues to write and ship packages spanning Python, Elixir, Zig, and Rust. Some of his recent open-source software libraries include:
- ExArrow: An Elixir wrapper for Apache Arrow data, allowing Arrow data to stay in native memory while giving Elixir opaque handles—vital for lightning-fast streaming and IPC data interchange in trading applications.
- ExZarr: An Elixir implementation of the Zarr storage format for large, chunked, N-dimensional numerical arrays (frequently used in AI and data analysis pipelines). [5, 6, 7, 8]

3. Scaling Engineering Teams & Safe AI

In his current role, he oversees AI and architecture for Finance IT, heavily focusing on the “safe and compliant” implementation of Large Language Models (LLMs) at enterprise scale. He frequently writes about system architecture, data sketches, cryptography, and engineering pragmatism on his technical blog. [4, 6, 9]

When evaluating modern AI coding productivity, Vassilakis has publicly advocated for robust language error messages and tooling. He warns against the hype of LLMs generating code in strict languages like Rust without proper oversight, noting that strict compilers often just shift failures from logic bugs to compilation errors without actually decreasing a team’s review burden. [10]

Would you like to dive deeper into any specific project, such as how ExArrow manages memory, or look closer at his written technical essays on financial data architectures?

Give me a summary of his written technical essays and the ideas he discusses

On his personal site and Medium profile, Thanos Vassilakis writes deep technical essays grouped around systems architecture, data optimization, and pragmatism in large organizations. Rather than focusing on trendy frameworks, his writings focus on foundational computer science concepts applied to real-world infrastructure. [1, 2]

The core ideas and themes running through his essays include:

1. Data Sketches & Memory-Efficient Algorithms

As someone managing massive data streams in trading and finance, Vassilakis frequently addresses the challenge of processing “big data” without crashing a system’s memory boundary. [2]

Probabilistic Data Structures: He explores the math and logic behind HyperLogLog, Count-min sketches, and Bloom filters. He explains how to count unique elements or track frequencies across millions of real-time transactions using fixed, negligible memory sizes rather than huge hash sets.
Apache Arrow Architecture: He writes extensively about building memory-efficient data pipes—such as native-memory data sharing across languages (like Elixir and Rust)—to prevent serialization bottlenecks.

2. Micro-Optimizations vs. Structural Bottlenecks

Vassilakis frequently writes about identifying what actually makes software performant, cautioning engineers against “premature optimization” of code syntax. [3]

Data Locality: His articles emphasize that in modern hardware, CPU cycles are cheap but memory access is expensive. He discusses writing cache-friendly code and structuring data linearly in memory to prevent the CPU from idling.
Pragmatic Benchmarking: He details how to profile systems correctly under load, illustrating how architectural flaws (like bad shared state or locking mechanisms) degrade performance much faster than a poorly written loop. [4]

3. Distributed Systems, Cryptography, & Actor Models

Drawing from his work building trading engines, his architecture essays focus heavily on maintaining high concurrency safely. [2]

Rejecting Global Mutexes: He advocates for architectures like the Erlang OTP Actor model (used in Elixir), arguing that modern scalable apps should abandon traditional MVC frameworks that rely on database locks or shared memory mutexes. Instead, systems should pass messages between isolated processes to handle distributed real-time updates natively.
Applied Cryptography: He discusses secure data-at-rest and in-transit protocols, emphasizing minimal, auditable cryptographic pipelines within enterprise systems. [2, 4]

4. Pragmatic Engineering Culture & Technology Selection

A strong thematic pillar of his writing is the human and organizational side of software engineering.

Rejecting “Language Bigotry”: Vassilakis writes critically about tribalism in coding. He champions selecting languages based strictly on their runtime properties (e.g., Elixir for massive real-time concurrency, Rust or Zig for safe, raw memory performance) rather than corporate or community hype.
The Scale Illusion: He analyzes the psychology of middle management in corporate tech, explaining how organizations often measure success by headcount and budget size, which inadvertently pushes teams to build over-engineered, overly complex software systems. [5]

Would you like a summary of a specific essay concept, such as how data sketches calculate cardinality, or how his team integrated Elixir with Apache Arrow?

What’s his views on AI transformation and big data

Thanos Vassilakis’s views on AI transformation and big data reject corporate hype in favor of mechanical sympathy, rigorous systems engineering, and math-driven data pragmatism. As an executive managing RBC Capital Markets’ Finance IT AI strategy, his perspective is heavily shaped by building scalable, production-grade financial infrastructure. [1]

1. View on Big Data: Architectural “Leaness” and Probabilistic Math

Vassilakis’s core philosophy on big data is that the industry is severely over-reliant on throwing raw computational power and memory at data problems. [2]

The Fallacy of Infinite Storage: He frequently critiques the approach of building massive, bloated data warehouses or data lakes that ingest unorganized, raw enterprise files. He argues that this results in catastrophic data lineage bloat and unscalable cluster costs. [2, 3]
Probabilistic over Exact Computation: He strongly advocates for using streaming, memory-efficient data structures—like HyperLogLog and Count-min sketches—to process real-time streams. His view is that for high-volume transactions, it is far more efficient to use deterministic math that gives a 99% accurate answer in fixed, negligible memory than it is to store and map terabytes of raw state.
Zero-Copy Data Pipelines: To safely manipulate large datasets across corporate domains, he advocates for technologies like Apache Arrow (which he wrapped for Elixir via his open-source ExArrow library). He believes modern big data engineering should focus on “zero-copy” in-memory data structures, allowing different programming languages to share data in native memory without serialization bottlenecks. [1]

2. View on AI Transformation: “From-Scratch” Grounding and Hype Skepticism

While heading enterprise AI transformation, Vassilakis is a vocal skeptic of treating modern artificial intelligence like a magic box. He emphasizes understanding the underlying software mechanisms. [1, 4]

Demystifying through Execution: Vassilakis practices what he preaches regarding AI grounding. In 2026, he launched Zynfer, an open-source, from-scratch LLM inference engine written entirely in the Zig programming language and targeted for specialized hardware like Apple Silicon and AMD GPUs. His approach highlights that true AI transformation requires engineers who understand how tokens are generated at the bare-metal GPU level, rather than just plugging into external black-box APIs. [5, 6]
The AI Coding Illusion: He has warned against assuming that AI-generated code will automatically make engineering teams faster. He points out that strict compilers (like Rust) don’t magically make LLM code secure; they often just turn logic bugs into compile-time errors. In his view, AI tools shift an engineer’s role from writing code to heavily auditing code, which doesn’t necessarily reduce the mental or regulatory burden in high-stakes environments like finance.
Corporate Metric Distortion: Vassilakis notes a major cultural roadblock to genuine AI and technical transformation: corporate middle-management structures. He argues that standard corporate cultures incentivize leaders to expand budgets and headcounts rather than streamline software. As a result, many “AI transformations” end up over-engineered to justify team sizes rather than being built lean and natively concurrent. [7]

Would you like to explore his technical implementation of AI, such as how his Zynfer engine handles LLM inference via Zig, or look into his specific critiques of enterprise data architectures?