Professional Work

A high-level overview of my last few years as a software developer.

Mozilla

• Firefox AI Runtime
  • Designed and built the Firefox AI runtime from scratch in Gecko.
  • Developed a model hub system for efficient model distribution and execution.
  • Helped extend the runtime to support native backends: ONNX Runtime (C++) and Llama.cpp.
• Team Building & Features
  • Ramped up a team of ML engineers to create features powered by the runtime:
    • Smart tab grouping
    • Search and suggest
    • Link preview
• Standards & APIs
  • Initiated the WebExtension AI API to expose AI-powered capabilities to extensions.
  • Represent Mozilla in the W3C Machine Learning Working Group, collaborating with other browser vendors to shape web AI standards.
• Cloud Inference
  • Designed and led the deployment of a litellm-based service for running inference in the cloud, complementing on-device inference and enabling scalable experimentation.

Also, check out the Mozilla blog and Mozilla Hacks posts by me for some of the work I do there. I try to write in them on a regular basis:

• Mozilla Blog
• Mozilla Hacks

Nuclia

• Contributed to NucliaDB, a vector search database optimized for AI-driven search and retrieval.

Elastic

• Principal Engineer → Tech Lead
  • Began as Principal Engineer for Automation & Tooling, then transitioned to Tech Lead for Ingestion features in Enterprise Search.
• Connectors Project
  • Founded and built the Elastic Connectors project from scratch.
  • Led a team to extend and maintain connectors, enabling integration of diverse data sources at scale.
  • The project remains in active use and continues to be a core part of Enterprise Search.

Side Projects

Outside of my professional roles, I explore building developer-focused tools that solve real-world problems.

Python-dev

I was part of the Python core development team, with a focus on packaging tooling such as distutils. During that time, I contributed to shaping several PEPs in the packaging ecosystem and developed features that landed in the Python standard library.

Although I eventually stepped back due to the lack of time, the experience taught me a great deal about large-scale open source collaboration and the challenges of balancing community needs with technical constraints.

Flake8

I am the original author of Flake8, a linting tool that combines PyFlakes and pep8 hence the name Flake8. I created it to streamline Python code quality checks into a single, easy-to-use tool.

I handed over maintenance of Flake8 many years ago, and the community has continued to evolve it. These days, I often recommend Ruff, which offers similar functionality with a much faster Rust-based implementation.

Molotov

One of my main focuses throughout my career has been building tools that help developers work more effectively, while ensuring they remain open source.

When we began developing web services at Mozilla, such as Firefox Sync, the scale we had to support was massive, so we needed a reliable way to run load tests.

Plenty of tools already existed, but none were simple enough to let us build and execute tests quickly. At first, we tried a “smart” approach: a central HTTP service coordinating agents that ran the tests. But it quickly became clear this didn’t scale well—you ended up fighting to scale the testing framework itself.
We eventually shifted to using cloud vendor orchestration services and existing dashboard systems for centralized result collection.

That experience shaped Molotov, a lightweight Python CLI designed with a KISS philosophy. It became just one thing: a single command to run load tests from a single host, nothing more.

I still use it from time to time today whenever I need to run performance tests on my personal and professional projects, and try to keep Molotov up-to-date.

Molotov Documentation