Learning area
First Principles Design
Suppose we want a model to help a team find pumps that need inspection. Choosing an algorithm leaves plenty unanswered. What can the team act on? Which readings arrive in time? Will the device and the people using it keep up? We will reason through those questions using physical inspection systems and software services, then build the calculations in Botie's workshops. Known relationships help us estimate a design; predictions about the world still need evidence.
Work out what the system needs
- Define the ML ProblemA model can produce a prediction without helping anyone make a better decision. We will start with Botie's inspection team, work out what they can act on, and decide what a learned model would need to improve. The information available at decision time will become part of the design.3 lessons
- Model the System Before Building ItA prediction service needs more than a fast model. Images must be prepared, requests can arrive together, and devices have limited memory. We will calculate where the work goes, then compare local and remote inference using the same task.4 lessons
Find out where the design holds up
- Compare Designs Under ConstraintsThe fastest candidate may miss too many faults, while the most accurate one may respond too late. We will identify which designs meet the requirements, examine the remaining trade-offs, and find the assumptions that could change our choice.2 lessons
- Test the Design and Plan for FailureA promising estimate needs evidence. Botie will compare prototype changes, route unusable inputs to a defined fallback, and bring the inspection service through a design review. Each passing test will have a clear meaning and a clear limit.3 lessons