Design Thinking for ML Systems at Scale

Applying design thinking principles helps solve human-centered problems when developing machine learning systems. The process involves several phases, starting with discovery to identify the true business pain point, followed by designing how a model fits into an existing pipeline. ML engineers must avoid common traps like misalignment between the machine learning problem and the business problem, or confusing the customer with the end-user. A single ML model, such as one for bounding box detection, can be applied in various ways like pre-labeling, active tooling, or quality assurance linting, each requiring different operational considerations. Building user empathy by directly observing or participating in tasks is critical for identifying horizontal issues and creating effective, reusable solutions.