Talk Python To Me #522: Data Sci Tips and Tricks from CodeCut.ai
69 snips
Oct 6, 2025 Quynh Truong, a developer advocate at Nixler and founder of CodeCut.ai, shares her wealth of knowledge on Python and data science. She delves into the art of crafting bite-sized tips, helping busy data scientists boost productivity. The discussion covers open-source tools, the balance between notebooks and scripts, and how to achieve reproducibility in workflows. Quynh also highlights the benefits of using Loguru for logging and Git for version control, making data projects more trustworthy and manageable.
AI Snips
Chapters
Books
Transcript
Episode notes
Adopt Fast, Compatible Package Tools
- Try using modern managers like pipx/uv to get faster installs while preserving pip compatibility.
- You can adopt uv pip install as a direct replacement to reduce friction without rewriting workflows.
Conda vs. Modern Dependency Tools
- Conda provides many prebuilt binaries but can be heavy and opaque about extra packages.
- UV/Poetry-style tools favor explicit, reproducible dependency graphs better suited for engineering handoffs.
Pin And Share Dependency Manifests
- Pin dependencies and keep an explicit manifest so experiments reproduce reliably across teams.
- Use a single tool to manage environments to simplify handoffs between data scientists and engineers.
