Learn & reason
Identify a model or policy with capabilities relevant to a defined physical task.
Technology
02 / CONEXIQFrom capable models.
To capable machines.
We’re investigating the engineering that connects AI capability to efficient, bounded execution in physical systems. Our technical direction spans the model, the runtime and the machine.
Shape & constrain
Explore how capability can be adapted to the task, device and execution boundaries.
Act & observe
Evaluate local behaviour, system state, failure conditions and recovery on the target platform.
Engineering principles
The system is
the unit of progress.
Progress depends on what the complete system can do within its real constraints.
- 01
A task is the starting point.
Define the useful behaviour, its operating conditions and how success will be measured before selecting the model or hardware.
- 02
Constraints belong in the architecture.
Memory, energy, timing, permitted actions and recovery are design inputs. They need to remain explicit as a system moves towards deployment.
- 03
The target device is part of the evaluation.
A capability should be evaluated where it is intended to run, including its interfaces, workloads and physical context.
- 04
Evidence travels with the implementation.
Configuration, assumptions, evaluation methods and observed behaviour should remain connected to the version being assessed.
Research becomes useful
through engineering.
Read the questions behind our technical direction and how we intend to evaluate progress.