2025-09-26 –, Nebula
It's easy to get into sticky situations when building language model powered applications. This talk helps you transform sticky messes into sweet successes.
This talk delivers practical insights from creating production-ready language model solutions with Python. Moving beyond theoretical possibilities, we'll explore the real architectural decisions, code patterns, and tradeoffs that determine success or failure. Through a progressive case study, we'll demonstrate how to effectively leverage Pydantic structured outputs, validation techniques, and thoughtful API integrations while avoiding common pitfalls. Perfect for Pythonic product teams looking to build applications that result in a sweet treat everyone will love.
While every product team faces pressure to integrate language models, translating API calls into reliable, production-grade features remains challenging. This talk cuts through the hype to deliver practical patterns for success based on real-world implementation experience, emphasising the value of maintaining product decision logs to track architectural choices and their downstream consequences.
The core of the talk focuses on implementing structured outputs with Pydantic as a bridge between unpredictable language model responses and structured Django backends. We'll cover prompt design for structured data, validation and regeneration strategies, data flow from language model through database to frontend, and approaches to tuning task accuracy for users. A live demonstration will showcase these patterns in a working application through a component-by-component tour.
We'll address balancing code quality with "vibe-coding" when working with language models, sharing strategies for maintaining engineering standards while allowing for creative experimentation. The talk tackles practical questions every product team faces: How much "AI magic" is appropriate? When does performance justify cost? How do you choose between reasoning models and flash models? And how do you validate that your language model components actually deliver value?
Attendees will leave with concrete architectural patterns, code examples for structured outputs, decision frameworks for balancing automation and performance, and strategies for managing complete data flows in language model applications. This isn't a theoretical exploration but a practical guide based on building features that deliver genuine value to users.
I am a Machine Learning Engineer with 4 years of Python and PyTorch development experience. As a tech lead, I've managed the project discussed in this talk, providing first-hand experience with the architectural decisions and implementation challenges of building production language model applications.
I've provided ML expertise to startups and the UK government, and I'm particularly interested in beneficial AI applications. My background is in Physics and Atmospheric Physics, where I interpreted large tropical cyclone datasets at Imperial College London.
My most recent three Python and PyCon talks are a language exploration trilogy, bringing together insights from Python, Mojo, and Rust into practical application development with language models:
PyCon/PyData DE 2025 (Darmstadt) - 🦀 Rüstzeit: Asynchronous Concurrency in Python & Rust
PyCon/PyData DE 2023 (Berlin) - Mojo 🔥 - Is it Python's faster cousin or just hype?
EuroPython Prague 2022 - 🐍 Large Language Model Zen
This talk returns to Language Model Zen three years on, showing how the theoretical possibilities discussed then have evolved into $10bn applications today.