Eric J Ma's Website

« 1 2 3 »

How to communicate with lab scientists (when you're the data person)

written by Eric J. Ma on 2025-08-24 | tags: biotech communication decisions statistics translation collaboration trust meetings probability stakeholders

In this blog post, I share practical strategies for data scientists and statisticians to communicate more effectively with lab scientists in biotech. Instead of overwhelming collaborators with methods, I explain how to focus on decision-making, translate complex analyses into actionable probabilities, and build trust through clarity. I also offer tips for structuring meetings and anticipating common questions. Want to know how to make your insights drive real decisions in the lab?

Read on... (3423 words, approximately 18 minutes reading time)
Wicked Python trickery - dynamically patch a Python function's source code at runtime

written by Eric J. Ma on 2025-08-23 | tags: python runtime llm security namespace compilation execution functions toolbot monkeypatching

In this blog post, I share how I discovered a powerful Python trick: dynamically changing a function's source code at runtime using the compile and exec functions. This technique enabled me to build more flexible AI bots, like ToolBot, that can generate and execute code with access to the current environment. While this opens up exciting possibilities for LLM-powered agents and generative UIs, it also raises serious security concerns. Curious how this hack can supercharge your AI projects—and what risks you should watch out for?

Read on... (2185 words, approximately 11 minutes reading time)
Data scientists aren't becoming obsolete in the LLM era

written by Eric J. Ma on 2025-08-15 | tags: productivity workflows evaluation metrics business science models ai tools measurement

In my latest post, I share how large language models are changing the data science landscape—not by replacing us, but by making us more effective and opening up new opportunities to build custom AI solutions. I discuss why our skills in measurement and evaluation are more valuable than ever. Curious how data scientists can thrive in the LLM era?

Read on... (914 words, approximately 5 minutes reading time)
Stop guessing at priors: R2D2's automated approach to Bayesian modeling

written by Eric J. Ma on 2025-08-06 | tags: bayesian variance r2d2 dirichlet multilevel glm regularization priors inference pymc

In this blog post, I share my journey exploring the R2D2 framework for Bayesian modeling, which lets you intuitively control model fit by placing a prior on R² instead of individual coefficients. I walk through its elegant extensions to generalized linear and multilevel models, showing how it automatically allocates explained variance and prevents overfitting. Curious how this approach can simplify your modeling and highlight the most important factors in your data?

Read on... (2318 words, approximately 12 minutes reading time)
From nerd-sniped to shipped using AI as a thinking tool

written by Eric J. Ma on 2025-07-21 | tags: automation ai memory design coding testing architecture prototyping review pairing

In this blog post, I share how months of hands-on struggle and learning paved the way for me to ship a complex graph-based memory feature for Llamabot in just two days—with AI as my design partner. I explain why you have to "earn your automation" and how AI can amplify, not replace, your critical thinking. Curious how pairing deep preparation with AI can supercharge your workflow and lead to breakthroughs?

Read on... (2456 words, approximately 13 minutes reading time)
How to use xarray for unified laboratory data storage

written by Eric J. Ma on 2025-07-15 | tags: xarray bioinformatics reproducibility cloud workflow alignment features laboratory datasets scaling

In this blog post, I share how using xarray can transform laboratory and machine learning data management by unifying everything—measurements, features, model outputs, and splits—into a single, coordinate-aligned dataset. This approach eliminates the hassle of index-matching across multiple files, reduces errors, and makes your workflow more reproducible and cloud-ready. Curious how this unified structure can simplify your experimental data analysis and save you time? Read on to find out!

Read on... (1487 words, approximately 8 minutes reading time)
Reflections on the SciPy 2025 Conference

written by Eric J. Ma on 2025-07-14 | tags: scipy python conference marimo tutorials llms xarray community networking career

In this blog post, I reflect on my 10th year at the SciPy Conference, sharing highlights from teaching tutorials, attending inspiring talks, recording informal podcast conversations, and contributing to open source projects. I discuss the power of community, the evolution of scientific notebooks, and the importance of financial aid in making SciPy accessible. Curious about the behind-the-scenes moments and lessons learned from a decade at SciPy?

Read on... (2556 words, approximately 13 minutes reading time)
Earn the privilege to use automation

written by Eric J. Ma on 2025-07-13 | tags: education assessment automation ai learning workplace skills process outcomes privilege

In this blog post, I reflect on the challenges of integrating AI into education and the workplace, sharing lessons from educators who found that unrestricted AI access can undermine true learning and assessment. I discuss why it's crucial to earn the privilege to use automation by first mastering foundational skills and demonstrating the ability to verify AI outputs. How can we ensure that AI enhances, rather than replaces, our critical thinking and problem-solving abilities?

Read on... (1218 words, approximately 7 minutes reading time)
The job your docs need to do

written by Eric J. Ma on 2025-07-07 | tags: documentation diataxis innovation tutorials guides reference ai strategy product jobs theory

In this blog post, I explore how combining the Diataxis documentation framework with Clayton Christensen's jobs-to-be-done theory can transform the way we write docs. By focusing on the specific outcomes readers want to achieve, we can make our documentation more useful and competitive—not just against other docs, but against all the ways people solve their problems. What happens when you treat your documentation as a product designed to help users get real jobs done?

Read on... (1381 words, approximately 7 minutes reading time)
One hour and eight minutes: Building a receipt scanner with the weirdest tech stack imaginable

written by Eric J. Ma on 2025-07-01 | tags: prototyping ai claude terminal fastapi htmx notion llamabot experimentation expenses

In this blog post, I share how I built a fully functional receipt scanning and expense tracking app in just over an hour using an unconventional tech stack—FastAPI, HTMX, Notion, and my own LlamaBot AI. I describe how Claude Code enabled a focused, terminal-based workflow that kept me in the zone and made rapid prototyping possible. Curious how combining unusual tools can unlock new possibilities and boost your productivity? Read on to find out!

Read on... (1154 words, approximately 6 minutes reading time)
« 1 2 3 »