Shortly after joining Flickr, we embarked on creating a new app called Yahoo! Photos (codename: Exposure)
This app truly felt like a startup, and in a way - it was. We were a small, tight-knit group coming up with original ideas and running with them.
While building this product from the ground up, we tried to incorporate ideas that could help other products in the company along the way. We wanted to build the features in a modular approach so that we could have layers of abstraction via SDKs.
Unfortunately, the core app was shut down when Flickr was nearing a deal to be sold. I had such a fantastic time creating this product with friends; it was truly some of the most enjoyable moments of my development career thus far. There are many features and tools that we built that I could talk about for hours.
What got me excited about this project was the machine learning tech we had from Flickr. The way it was able to sort and categorize photos was unlike anything I had seen thus far in photo technology. I loved seeing the reactions of friends when showing them features we were building.
Here are a few of the features that I think are worthy of writing about:
I believe this is the code you’d scan to add Barack Obama as a friend :)
We implemented our own social graph using a simple Firebase database (like a true startup).
Then, to expand on this, we worked with the vision team to create a feature that would generate a custom image that could identify a user based on their unique id. To add your friend in the app in a fun, sure-fire way would be to simply scan their code. This is pretty common in social apps today (2020).
Shareback Feature - “When a photo is shared with a user on a Yahoo platform, Shareback aims to provide that user with suggestions to share back based on metadata of the photo sent and of the recipient's photo library.”
Let’s say you go on a trip with your friends and take photos. If one of your friends sends you a photo that is near the same location and timestamp as any of your photos, we’ll automatically stack them together and suggest sending them back to your friend. We achieved this by creating a leader-follower clustering. Once clustered, we compute the likelihood of a location being Frequent, Familiar, Away, and Rarely visited by the photographer as to not share common, irrelevant photos. Additionally, we tried to have a smart approach to removing blurry photos, accidental bursts, and near duplicates. Another moment that felt like magic was adding Siri support. Being able to say “Send my latest uploads from Yosemite to John” and have it *just work* was incredible. There were definitely edge cases and nuances when working with Siri that we did our best to work around.
UX flow of digitizing physical coupons.
As the core app was nearing the end, we wanted to extract as much value from it as possible. There was an idea to automatically detect if a user has a coupon in their camera roll and to enable the user to deliberately take photos of their coupons to digitize them. If the user was near a store they had a coupon for, we would send them a push notification. This was fun tech that after the initial version was completely redone by the core Yahoo Mail team. Remnants of this are in the “Deals” tab in Mail today.
Helping create the machine learning model to detect coupons. This was a fun break from normal work. The entire team would bring in coupons to be categorized and later be used to create a machine learning model.