Facial Data Collection

What I did:

I was brought in to fix a facial expression data collection study after an initial 15-person pilot yielded unusable results for the ML Vision team due to poor experimental design. I redesigned the research protocol to be human-centered, ensuring technical requirements aligned with natural user behavior. This new blueprint was validated through a 400-user study that delivered the high-precision data the ML team required. When the team expanded the scope to include additional movement modalities, I revised the UI and instructions to maintain data integrity. The resulting blueprint was adopted as the standard roadmap for hundreds of subsequent global recording sessions.
 

Methods:

• Dogfooding: Personally underwent the initial ML study protocol to identify ergonomic "walls," cognitive friction points, and instructional gaps

• Heuristic Analysis & Iteration: Redesigned the study order and UI instructions to prioritize human-centered flow and naturalistic movement.

• Internal Validation: Conducted in-house testing with small cohorts to verify data consistency and participant adherence before global deployment

• Global Protocol Deployment: Scaled the validated blueprint to international recording studios for high-fidelity data collection across hundreds of sessions