Joe Little

I've worked on c++ trading applications for over 10 years and have extensive experience with most features (and many extensions) of the language - OOP, templates, macros, STL, boost, debugging with gdb and sanitized builds, performance analysis and optifmization, and more. Maintaining low latency in our critical path is always a constraint, though I haven't focused primarily on extreme high performance in recent years. I am a c++ generalist who can build a c++ application to do nearly anything within the scope of the language. I try to write code that can be easily understood, extended, and modified, with appropriate levels of abstraction. Getting it done is most important but clear and extensible code will make or break a system's long term success. I use c++ over other languages for performance-critical tasks where nothing else will do. As a side note, over the years I've also embedded c++ modules into systems built mainly in another language, to combine the benefits of c++ performance with the fast development, user-friendliness, and libraries of other languages - e.g unmanaged c++ in a c# context, or c++ with python bindings via boost.

I've used Python in many contexts, such as: 1. Dev ops automation - process monitoring, automated regression testing, build deployment, web scraping, and much more 2. Code generation (c++) based on XML message specs 3. Binary data parsing with struct.pack 4. Data generation (e.g. CSV manipulation) 5. Code inspection / AST manipulation for optimization purposes 6. Launching and monitoring thousands of parallel jobs and then combining the results (similar to hadoop but via qsub) 7. Real-time data dissemination 8. Database interaction (various implementations of SQL) Python is a huge, flexible language with so many modules. I view it as a general purpose language, and while it is not always the very best for a particular task, it can be used for nearly anything.

I use pandas and XArray to manipulate financial data, for example stock-time indexed data where we need to tweak, clean, combine, aggregate, and perform transformations to generate or calibrate trading signals or provide clean input to other systems. Often this data is very large so memory usage and performance become important constraints. I've found that choosing how to index data, when to use low level code (numpy or cython) vs built-in dataframe or xarray functions, and sometimes the selection of non-obvious pandas function kwargs, can have orders of magnitutde implications for how the code performs with particular datasets. There are often many ways of doing things in the pandas world and I enjoy investigating and testing approaches and figuring out the best one.

I've worked on real-time automated trading systems for my entire career (going on 12 years now), and I always enjoy putting out the occasional fires that come up. Being good at dev ops requires understanding the system or at least the technology, knowing the pieces involved and how they interact, and being able to look at code and quickly understand what it is doing. It is also a science of logical deduction, process of elimination, and hypothsis testing, and this is the part that I enjoy the most. Usually when a problem comes up there is an investigative phase where you try to figure out exactly what is causing the problem, sometimes which line of code is an issue, or which input triggered the problem. A good dev ops person will have some hypotheses or candidate components or inputs and then will devise in real time tests or checks that eliminate or confirm possible explanations, weighing the time used and risk of performing each check against the potential information gain. Often it resembles a binary search of explanations until the problem is understood. I love the moment when a right-feeling hypothesis occurs to me, but even better is when after resisting the temptation to declare it as the correct answer, I can finally prove it. Mitigating actions may be taken before and/or after the investigation, weighing the risks and benefits and utilizing an undestanding of the system to know what is possible and what the consequences of potential actions might be. Often this requires some creative thinking, and I enjoy thinking through the possibilities and their implications. It requires good foresight and decision making. It can be a lot of fun. The cure for dev ops banality is to automate solutions and monitoring for problems that come up repeatedly (or ideally the first time they come up) both to reduce the risk of issues and to free up dev ops to focus on novel problems. I enjoy the process of creative automation. I'd love the opportunity to help you put out your fire!