The Human Code
A working roundtable for developers building beyond correlation-based personalisation.
LLMs understand language remarkably well. Understanding an individual is a different problem.
Most personalisation systems still rely on patterns observed across a population and map those patterns onto the person in front of the model. That works—until shifting intent, conflicting behaviour, changing context, or individual causality breaks the assumption.
The Human Code brings together 30 developers, ML engineers, researchers, and technical founders who have encountered that ceiling and want to work through it together.
No stage. No keynote. No sales pitch. No product showcase.
Bring one real problem. The room will work on it with you.
What to expect
5:00–5:15 PM — Arrivals
Check in, meet the room, and settle in. We begin the working session at 5:15 PM sharp.
5:15–6:00 PM — Live model comparison
We will run an LLM and an LBM—Large Behavioural Model—on the same individual-level inference task.
This is not a polished demo reel. It is an open working session.
Participants can challenge the assumptions, introduce new inputs, question the task design, and attempt to break both approaches. If the room believes the example is too convenient, we will use a participant’s problem instead.
6:00–7:15 PM — The working roundtable
Every participant gets the floor.
Bring one specific use case where personalisation, user modelling, or individual-level inference is failing.
For example:
“How do I model shifting intent when recent behaviour contradicts the user profile?”
is more useful than:
“How should I improve personalisation?”
The room will examine each problem together. Participants working with causal methods, graphs, reinforcement learning, state-based models, memory systems, embeddings, or other approaches will have equal space to contribute.
This is a shared blackboard—not a Dots-In showcase.
7:15–8:00 PM — Patterns, disagreements and open conversation
We will identify the