Alexander Ocsa

System Software Engineer

Systems engineer with a PhD in computer science and over a decade of hands-on experience accelerating analytics, I've zeroed in on GPU-accelerated query engines and high-performance numerical computing stacks. Throughout my career, I've crafted sophisticated parallel software that taps into the raw power of diverse hardware accelerators—like GPUs, pushing them to their limits in real-world applications. I'm excited to dive in and elevate the functional depth, raw speed, scalability, and rock-solid reliability of those essential data processing operators at the heart of analytics engines or deep learning stacks.

About

I have actively contributed to several significant open-source projects:

  • PyTorch: 40+ PRs in Aten / scatter-tensor kernels, numerical accuracy perf fixes, code reviews, and design discussions that improved reliability and throughput for large workloads.
  • Apache Arrow: +9 PRs Improvements in performance, new features, testing integration, and fixes that enhance ecosystem interoperability.
  • RAPIDS/CUDF: +7 PRs in CUDA primitives and analytics kernels (groupby/join), plus stability and testing improvements used in production analytics.

Work Experience

VoltronData

Remote
06/2021 — present

Staff Software Engineer

  • Designed physical execution abstractions with spill paths across GPU HBM → RAM → Disk, enabling stable execution of memory-intensive queries.
  • Shaped a distributed count-distinct algorithm that delivered ~5× speedup vs. prior approach at cluster scale.
  • Championed best practices for code-splitting strategies introducing leak-spotting structures tuned to join types; improved reliability and tail latencies.
  • Enhanced the Apache Arrow codebase by contributing to both C++ and Python, while developing clear, comprehensive documentation to streamline onboarding and support future contributors.
  • Proposed and implemented core architectural improvements to the Theseus engine, contributing to scalability and performance enhancements.

Quansight

Remote
05/2020 — 05/2021

Senior C++ Software Engineer

  • Shipped advanced numerical and deep-learning features in PyTorch, focusing on high-performance kernels and correctness at scale.
  • Implemented new functionality and optimizations in C++/CUDA/Python, improving throughput and reliability across core components used by a large global user base.
  • Built GPU-accelerated sparse-tensor kernels and infrastructure enabling efficient processing of large sparse datasets and removing critical bottlenecks.
  • Optimized low-level ATen operators and execution paths, eliminating inefficient memory/compute patterns and improving overall performance characteristics.
  • Collaborated in public issues/PRs with maintainers and users to land changes with strong tests and documentation.

BlazingDB Inc.

Lima, Peru
02/2018 — 05/2020

Senior Software Engineer

  • Designed the original execution model combining three async executors—compute (libcuDF), comms (UCX/TCP), and memory/spill (host/disk)—to improve backpressure handling, utilization, and resilience under memory pressure.
  • Built a UCX-based multipart GPU messaging layer, increasing network throughput and reducing CPU overhead for distributed queries.
  • Led the physical plan + cache layers of a pull-based execution engine, enabling efficient processing of memory-intensive queries and higher sustained throughput.
  • Detected pre-GA defects ~30% by overhauling unit/integration tests and CI signal, reducing first-release stability and user experience.
  • Profiled end-to-end system trade-offs (I/O, messaging, kernel concurrency) and tuned policies/parameters to unlock better multi-node scalability.

Laboratoria

Lima, Peru
06/2017 — 02/2018

Software Engineer & Bootcamp Teacher

  • Led a teaching and instructional role as a Bootcamp teacher, guiding and developing talent through comprehensive e-learning platforms and hands-on training.
  • Focused on building skills growth and career readiness in aspiring software developers.
  • Led the development of talento.laboratoria.la, a networking platform that connected over 300 companies with bootcamp graduates, contributing to 1,000+ successful employment matches. Played a key role in both front-end and back-end development to ensure a seamless user experience and effective platform functionality.

Northern World Entertainment Software Inc

Toronto, Canada
05/2015 — 11/2016

Software Engineer

  • Contributed to the development of cutting-edge mobile and game technologies, focusing on Holographic Mixed Reality Apps for the Microsoft HoloLens.
  • Successfully launched these applications on both the iOS and Android platforms, enhancing user experiences through immersive and innovative technology solutions.

Education

PhD in Computer Science

Arequipa, Peru
2013 — 2017

Universidad Nacional de San Agustin

Public congratulations for PhD thesis (Summa cum laude)

M.Cs in Computer Science and Mathematics

Sao Paulo, Brazil
2010 — 2012

Sao Paulo University

Honors: Maximum Distinction

B.S. in engineering with a major in Computer Science

Arequipa, Peru
2004 — 2009

Universidad Nacional de San Agustin

Ranked 1st position in Engineering (class 2004-2009)

Skills

C++ CUDA Python PyTorch Apache Arrow RAPIDS/cuDF GPU Computing Query Engines Distributed Systems Performance Optimization Numerical Computing Deep Learning Algorithm Design Database Systems

Featured Project Contributions

GPU Query Engine (Theseus)

2021 — Present

Building a petabyte-scale GPU-accelerated query engine with advanced memory management and distributed execution capabilities.

C++ CUDA Apache Arrow

PyTorch Sparse Tensors

2020 — 2021

Enhanced PyTorch's sparse tensor capabilities with GPU-accelerated kernels for efficient large-scale sparse dataset processing.

C++ CUDA Python

BlazingSQL GPU Query Engine

2018 — 2020

Designed and implemented the distributed GPU SQL engine with UCX-based messaging and pull-based execution model.

C++ CUDA UCX

Let's Connect

I'm deeply passionate about data structures and algorithms, especially how they evolve and scale. I enjoy solving the complex puzzles of system optimization, where each improvement brings new possibilities and challenges.