Bosch

I analysed end-of-line production test data from radar sensor manufacturing with Python and SQL under Git version control, building machine learning models and supporting 8D root-cause investigations when parts failed, looking for the parameter drift or test-limit pattern behind a defect. The engineers who owned each investigation reviewed the work, and the findings went back to them as evidence, so decisions about the sensors rested on measurement rather than on assumption.

8D is the automotive standard for working a defect back to its cause, and the data work sits in its fourth stage. That stage asks two questions. What caused the fault, and why did the production line let it through? The first is answered by comparing the measurements of failing units against passing ones until a parameter shows drift. The second is harder, and the answer is usually uncomfortable: the test that should have caught it was set too loosely.

An analysis nobody opens changes nothing. The findings went into charts and dashboards the engineering teams could read without me in the room, which meant working out what a quality engineer actually needs to see: the parameter, the limit, and where the distribution sits against it.