Eric J Ma's Website

I've just explored a fascinating paper on differentiable search of evolutionary trees. It's a creative blend of math and biology, using mathematical structures to solve a biological problem. The authors have developed a way to infer both phylogenetic trees and ancestral protein sequences in a continuous, differentiable manner. This opens up exciting new avenues for protein engineering and design. Plus, the paper's figures are top-notch! 🧬🌳📊

In this post, I discuss the 'Wayne Gretzky move' in tech adoption, especially in data science. It's about foreseeing the future of tech and making bold moves to stay ahead of the game. It's risky but with the right team, it can lead to significant gains. 🏒💻🚀

In this post, I discuss how to decide which technological models to adopt in our work. I introduce two frameworks: the 'technological trinity' and the 'technology ladder'. The former helps us evaluate if a new technology is worth investing in, while the latter places it within a broader context. I illustrate these concepts using protein engineering as an example.

Just built LlamaBot, a chatbot using OpenAI GPT-4, and it's been a wild ride! 🎢 From dealing with rapid innovation and versioning issues to discovering the power of code ghostwriting, it's been a learning curve. 🧠💡 Also, I explored some cool LLM frameworks. Dive in to learn more! 🏊‍♂️

In today's blog post, we dive into setting the minimum and maximum Python versions in pyproject.toml 🐍. We explore how this impacts llamabot and discuss the new syntax for indicating the union of types in Python 3.10. This new syntax is visually easier to understand, making our coding journey a bit smoother! 🚀👩‍💻👨‍💻

Tired of writing boring commit messages? 😴 I've got you covered! Introducing llamabot, a CLI tool that uses GPT-4 to craft meaningful commit messages following the Conventional Commits specification. 🚀 Improve collaboration, project management, and debugging with just one command! 🌟 Try it out with pip install llamabot. 🦙

Today, I explored the outlines library and built a coding bot that ghostwrites code, docstrings, and tests! 🤖 With clean prompt design, I was able to create a compact and efficient Python program. The best part? It's super easy to map this to a user interface! 🎉 Check out how I did it and learn about the benefits of using outlines for prompt management. 😄

Today, I discovered a fantastic use for GPT-4: writing improved pull request messages! 🎉 I used a Python package I developed, llamabot, to generate summaries for pull requests. The results were impressive, making the task of reviewing PRs with long diffs much less daunting. 😌

I’ve been working on llamabot and wanted to implement a feature to automatically record prompts and LLM responses. I used a PromptRecorder object with a context manager, similar to PyMC’s pm.Model(). However, I faced a challenge: making the context manager aware of its internal code. 🤔 After some research, I found a solution using the built-in contextvars module. By creating a globally-referenceable prompt_recorder variable, I was able to modify the instantiated PromptRecorder object’s state and call its methods while keeping it hidden from the front-end code. 🎉 Now, llamabot can easily record prompt-response pairs! 🤖

🧐I encountered a tricky problem when working on chromatography: I needed to find x-values on concave curves at a specific height without solving simultaneous equations. Check out my blog post 📝 to see how I went from a simple for-loop solution to using JAX's vmap for a more effective, expressive, and compatible solution. This was a fun programming problem to solve! 🔎📈📊😃