A custom Microsoft Power Platform application — built in PowerApps, orchestrated by Power Automate, backed by SharePoint — that digitized quality control across the company. Field techs run 11-point and outbound inspections from their phones; flows automatically store, categorize, and clean the results. That clean, structured data is exactly what feeds the RTU QA vision model.
My title is quality control inspector, but I treated the role as a mandate to improve the process itself. The old Microsoft Forms flow took 1 minute 40 seconds per inspection. Knowing that 90%+ of units pass, I designed the app to pre-fill a passing result and let inspectors focus on triaging the exceptions — dropping average entry to 22 seconds, an ~80% cut that recovers 40+ hours a year across ~2,000 inspections. Deploying it meant writing the documentation, training the team at each centre, and earning adoption from three groups that each had their own way of working.
Run through the flow the way the field does it: complete the PowerApps inspection on the left, submit, and watch Power Automate route the record — storing the photo, categorizing the QC result, cleaning the fields, and landing it all in SharePoint, ready to feed the model.
The cleaning step is where the value compounds: normalizing serials, de-duplicating captures, and tagging pass/fail is exactly what turns messy field photos into a labelable dataset — the bottleneck the RTU QA project lives or dies by.
Most ML failures are data failures. This project shows I can meet an organization where it already lives — inside the Microsoft 365 stack — and build the durable, automated data plumbing that makes downstream AI possible. It's the same instinct that runs through every project here: the model is only as good as the pipeline feeding it.
See where the data lands: RTU QA →