Oxxo Corner / Femsa

Context

Oxxo Corner explored a different way to interact with an Oxxo store: use the phone’s camera and location context to discover products and recommendations. The first version combined an augmented-reality experience, image filtering with generative AI, product information and a community forum.

My role

I was the Product Owner and a full-stack developer on a team of four. I turned the idea into a first-version scope, coordinated the SwiftUI screens with the Flask and Python API, and kept the AR navigation, profile and community flows connected rather than treating them as separate demos.

Decisions

• Use ARKit for directions, not as decoration. Store locations and product cues had to answer a shopper’s next question. The trade-off was device-specific testing and more careful fallback behavior when tracking was weak.
• Keep CoreML image recognition close to the iOS client. On-device inference can respond without sending every camera frame to the backend, but it adds model-size and device-compatibility constraints.
• Put accounts and community data behind Flask. The service gave profiles and forum interactions a shared persistence boundary. That costs API and permission work, yet keeps the mobile client from owning business data.
• Limit generative AI to image filtering in the first version. A narrow use kept the experiment understandable and reduced the risk of making product recommendations sound more certain than the inputs allowed.

Result

The team produced a coherent first version that connected store discovery, AR directions, product recognition, recommendations, profiles and community discussion in one iOS flow.

What I would do differently

I would validate AR navigation with real store layouts before expanding the social layer. That would reveal tracking and signage constraints early and give the team a clearer basis for deciding which parts of the concept deserved another iteration.