Building Data Science with Foundation LLM Models
Talk Python with Michael Kennedy and guest Hugo Bowne-Anderson
Large language models are starting to change how data science is done every day. They can help write and review code in notebooks, assist with data preparation, and power new types of workflows. But turning these capabilities into real AI products requires more than prompt engineering. It requires evaluation, iteration, and operational discipline.
In this Talk Python live recording, Hugo Bowne-Anderson, data and AI scientist, educator, podcaster, and longtime data science community builder, joins us to break down what it takes to ship with foundation models. We will explore the skills, practices, and tools that help data scientists build reliable LLM-powered applications and move beyond proof of concept purgatory.
What we will cover together
Where LLMs actually help data scientists and where they collapse into demo ware
How to avoid proof of concept purgatory and the common failure modes that kill projects before launch
The evaluation loop that separates vibes from production
How to choose between RAG and fine tuning without wasting months
How to make flip floppy models stable enough to ship
How to build and evaluate agents and where agentic value modes create real value
What production realities look like including cost, latency, monitoring, and serving
What skills matter most for careers in the LLM era and what is over hyped going into 2025
Why this conversation matters
Too many AI talks stop at shiny prototypes. This one is about building real AI products. You will leave with a clearer view of how evaluation, iteration, and data science fundamentals come together to make LLM applications reliable. You will also get perspective on how to grow your own practice and career in this new era.
If you can’t make it, register and we’ll share the recording after.