Jan 2024 – Present · 2 yrs 4 mos · Remote

• My work on Apache Parquet focusses mostly on the C++ implementation, which is tied to Apache Arrow. I also contributed a specification improvement (broadening supported data types for a compression-friendly data encoding) and the corresponding C++ and Java implementations.

May 2021 – Oct 2024 · 3 yrs 5 mos · Remote

• I provided technical expertise, guidance and leadership to a team of talented people who worked on Apache Arrow and related technologies.
• I made numerous contributions to the Apache Arrow and Apache Parquet open source projects (see related experience entries).

Jan 2019 – Apr 2021 · 2 yrs 3 mos

Jan 2018 – Present · 8 yrs 4 mos · Remote

• My work on Apache Arrow covers multiple areas. Initially, I hardened and expanded the foundations of the C++ and Python implementations, such as: setting up fuzz testing for critical components, improving the IO facilities, designing a filesystem abstraction, making numerous quality of life improvements, writing our initial prose-based documentation. I also helped improving the usefulness and applicability of Arrow through standardization. For example, I designed and implemented the C Data Interface, a language-agnostic ABI-stable protocol for zero-copy data sharing, and its iterative counterpart the C Stream Interface.
• Finally I play a major role in the community, participating in design discussions, making numerous code reviews, ensuring a consistently high code quality, helping contributors towards completion of their work.

Sep 2014 – Dec 2017 · 3 yrs 3 mos · Paris Area, France

• I spend two years working on Numba, a Just-In-Time compiler for Python with specializations for scientific computing. Areas of involvement include implementation of compiler optimizations and maintenance of an LLVM IR generation library (llvmlite).
• Then I work on Dask / distributed, a framework for concurrent and distributed data analysis in Python. I specialize on low-level aspects such as network programming and the distributed scheduler.

Jan 2012 – Jan 2013 · 1 yr · Paris Area, France

• I lead our small R&D team, setup development and quality practices and tools, bump our test coverage ratio to a whopping 75% (and counting), rearchitect critical parts of the software stack, strength and clarify internal APIs, and of course contribute the required new features and improvements. In other words, "getting things done (and mostly right)" is the motto. Also, I have a say in product development decisions and planning.

Jan 2008 – Present · 18 yrs 4 mos · Remote

• I was given core developer access after spending a few months contributing bug fixes, improvements and reviews on the Python bug tracker.
• Some notable aspects of my work on the Python interpreter:
• I rewrote the Python 3.0 I/O stack in C, yielding massive improvements in speed (as much as 18x)
• I helped implement a new datatype (the memoryview object)
• I contributed various improvements in performance (e.g.: bytecode evaluation loop, garbage collection heuristics)
• I completed the implementation of some delicate parts of the Python 3.0 exception handling semantics (lexically scoped exception state handling, chained tracebacks)
• I re-implemented the Global Interpreter Lock to minimize contention overhead and make thread switching more predictable
• I improved garbage collection semantics to solve the finalization problem in reference cycles (PEP 442)
• I maintained and massively improved the ssl module over the duration of several years, before finally giving the reins to people more qualified than me
• Here is a list of PEPs (Python Enhancement Proposals) I wrote or co-authored :
• PEP 385: Migrating from Subversion to Mercurial (accepted)
• PEP 407: New release cycle and introducing long-term support versions (rejected)
• PEP 428: The pathlib module -- object-oriented filesystem paths (accepted)
• PEP 442: Safe object finalization (accepted)
• PEP 455: Adding a key-transforming dictionary to collections (rejected)
• PEP 556: Threaded garbage collection
• PEP 3151: Reworking the OS and IO exception hierarchy (accepted)
• PEP 3154: Pickle protocol version 4 (accepted)
• PEP 3155: Qualified name for classes and functions (accepted)
• I also spend much of my time reviewing and assessing other people's patches, especially newer contributors.

Sep 2006 – Jun 2007 · 9 mos

• Internal quality audit on the WengoPhone project.
• R&D, design and implementation of disruptive new services (WengoVisio, WengoMeeting and more to follow).
• Implementation of an RTMP server using Python, Twisted and gstreamer.
• Implementation of a SIP stack using Python, Twisted and gstreamer.
• Prospective work on Jabber and SASL.

Nov 2004 – Jan 2006 · 1 yr 2 mos

• Lead engineer of the Solipsis research project, a peer-to-peer protocol for decentralized virtual worlds. Implemented and evolved the P2P protocol, coordinated the open source community.
• More information at http://solipsis.netofpeers.net/

Jan 2001 – Jan 2003 · 2 yrs

• Designer and implementer on an internal project to monitor routing protocols, so as to detect and qualify network incidents and ultimately improve quality of service. The software captured BGP and IS-IS packets in real time and stored advanced information in a database so as to make various statistics and diagnostics available on an interactive Web interface. It has been used in production for several years with much success.

Sep 2001 – Jan 2012 · 10 yrs 4 mos

Jan 1999 – Jan 2001 · 2 yrs

• Design and prototyping of interactive TV applications.

Jan 1998 – Jan 1999 · 1 yr

• I designed and started prototyping the artificial intelligence of a first-person shooter game for Windows PCs. After single-player gaming was cancelled in favour of multi-player gaming, I proceeded to work on various remaining parts of the project. Finally, the whole project was abandoned.