Python Glasgow

Ever wondered how ChatGPT remembers your name, your job, and that project you mentioned three weeks ago? The model itself can't. Underneath, every LLM call starts from a blank slate. The memory is a separate system someone had to design and build, and when you build your own AI agent in Python, that someone is you.

It's a more interesting problem than it looks. You can't just save everything, or the agent drowns in its own history. You need short-term memory for the conversation you're in and long-term memory for everything that should survive after the tab closes, and they demand completely different designs. You need write policies to decide what's worth keeping, semantic search to find "hates early flights" when someone asks about a 6am departure, and rules for when something said in March should override something said in January.

In this talk we'll build a Scottish travel agent (in Python) that remembers your preferences, your budget, and your dodgy knees across conversations. We'll plan some trips, then open up its memory and look at what it actually wrote down: what the user told it, what it merely inferred, the moments it confuses the two, and how each mistake gets fixed.
No AI experience needed. You'll leave with a working repo to play with, a mental model for how memory works in every AI product you use, and a checklist for the memory system you've probably already accidentally built.

🎯 Target Audience

• Python developers building AI agents
• AI/LLM practitioners who want to understand memory
• Developers new to AI who want a practical introduction
• Product/AI engineers designing conversational systems