Building Tomoro x AI Demo Night: Getting AI to Production While the Ground Keeps Moving
Most teams are still catching up to last week's release instead of adapting.
This invite-only evening is for engineers and product leads building real systems in real organisations, where the models keep changing, the enterprise won't, and shipping still has to happen on time.
We're looking at two production systems we've built: one that had to wrestle reliable, structured outputs from the chaos of enterprise data, and one that had to be continuously adapt as the AI landscape shifted beneath it.
Together, they answer the same question from different angles: how do you build for production when the ground keeps moving?
We’ll follow these presentations with the usual demos night. If you’re a builder, you get to show what you’ve been creating. This doesn’t need to be polished. If anything, we want scrappy and interesting. And remember show us the thing, not the slides. Submit your demo idea below for approval.
Agenda:
🍕🥤🤝 6pm arrival for pizza, drinks and a friendly chat
📢 6:45pm - Building Tomoro sessions begin (info below)
⏱️ 7:15 - Break
⚒️ 7:25 - Four demos
Building Fact Extraction Systems in Enterprise Environments — Lloyd Hamilton
Enterprise data is messy and LLMs are non-deterministic. But production systems we build have to be neither.
We'll walk through the common paradigms for fact extraction at scale and where each one breaks down in practice. We'll cover the techniques that actually tame LLM variability in structured output tasks, and be honest about why getting any of this across the line in a large organisation is a problem that goes well beyond the model.
How We Built the Virgin Atlantic Concierge Against the Pace of Change — Eilidh McMenemie
The technology landscape continued to shift during our build. We started building against one set of tools as better ones arrived. A new Realtime API, MCP, the GPT-5 suite. Ignoring them wasn't an option but neither was rebuilding from scratch.
This is the story of how we architected a production AI system to stay adaptable: shipping continuously while absorbing new capabilities as they landed, without losing the thread of what we were actually building for.