The opportunity wasn't simply to improve the model. It was to improve the system around it.
MVP Index used AI-based logo and image detection to measure sponsorship exposure. Model creation and correction depended on data labeling, quality controls, supporting tooling, and coordination across technical and operational teams.
The bottleneck wasn't isolated to model behavior. It lived in the workflow connecting people, data, tooling, evaluation, and correction.
I translated customer, operational, and technical problems into product requirements, prioritized roadmaps, backlogs, dependencies, milestones, and measurable outcomes.
Working with Engineering, Data Science, QA, and Operations, we redesigned data-labeling and model-correction workflows, strengthened quality controls and supporting tooling, and established evaluation practices, human-in-the-loop workflows, documentation standards, guardrails, and reusable processes.
Model performance improved from roughly 50–60% accuracy to approximately 95%, while the surrounding evaluation and operating practices made correction and iteration more repeatable.