Wasn't our final project but we developed an edge based face recognition system. You could upload a photo of a person of interest and train a face recognition model with it. This model was then distributed to our edge devices (a Rapsberry Pie with a webcam).
Unlike most face recognition systems we leveraged fog computing, meaning that the face recognition happened on the edge device itself. This means that data was being transferred to the backend only in case a match was found, thus providing better privacy.
We also added things like immutable logs and four-eyes principle to every action in order to facilitate accountability.
Backend: AWS IoT, DynamoDB, Lambda