Open Source in Finance

Open source in finance probably sounded absurd a decade or two ago - why would big banks share anything, right? But now I think the composition and culture of the financial industry is changing. From anecdotal and personal experience, new hires at top hedge funds, prop trading firms, and even prestigious banks tend to be IMO, Putnam, and IOI winners. These intellectual athletes are deeply embedded in pure math, computer science, and AI and are much more in touch with the modern software world. These firms are realizing they can’t afford to build every piece of software themselves, especially when open source tools like Pandas, Apache Arrow, or even Kubernetes are already battle-tested and widely used. What’s even cooler is that some banks are now contributing back—Goldman Sachs open-sourced their modeling platform Legend, and JPMorgan has done similar things with their visualization tools. It’s part talent strategy, part reputation play, and part just keeping up with the speed of modern infrastructure. The finance world is still pretty secretive, but open source is pushing even the most traditional institutions to rethink how they collaborate and build.

Open Source AI

Nick Vidal’s presentation tried to define what constitutes open source in AI. Admittedly, I have almost no background in AI, so take the following opinion with a grain of salt. AI should absolutely be open source—but that’s only part of the solution. The real issue is that companies like OpenAI and Anthropic have billions in compute and data resources, which gives them a massive head start that no open community can realistically match. Even if the code or weights are released, there’s still no real way to verify what data was used, what biases may be baked in, or how the training process was conducted without that same scale of infrastructure. It’s like being handed a finished dish with no recipe—transparency isn’t just about the final product, it’s about being able to reproduce and audit every step. Without open access to the full pipeline—data, compute, methodology—”open source” alone risks becoming a hollow label.