August 16, 2019
Categorized: IBM News | IBM Today | IBMer Stories | Students
Share this post:
(3 min read) By Joel Joseph
I’m a huge dog lover, like the biggest dog person you’ll ever meet. And as a dog person, the biggest problem for me is that I have no dogs at home. That’s why I decided to go to the Humane Society this summer and see how hard it would be to possibly adopt one.
One major thing that working on startups has ingrained in me is the lean startup methodology, specifically when it comes to customer discovery. And it has led me to always want to understand more about everything around me and how it works. I’m not specifically looking for problems, more like stumbling across them.
In the process of understanding how adoptions work, I had the opportunity to talk to six different staff and volunteers and casually ask them about their jobs and experiences. As I spent more time talking to them, each spoke about different problems and pain points in the adoption process. At the end of the day, I not only walked away with a huge amount of respect for the volunteers and staff but with a fundamental problem that shelters were facing: no one knew what types of dogs they were giving to families.
Me with one of the volunteers at my local shelter
I learned that a lot of the dogs that come through my local shelter are from rural communities, which many times don’t have a lot of history associated with them. When the shelter takes a dog in, they have a staff member basically “eyeball” what type of dog they are dealing with. As a result, there is no way to actually know which type of dog people are adopting. Now one common solution to this could be genetic testing each dog that comes through, but at 100 dollars a genetic test per dog, it’s a non-starter for a lot of nonprofits who don’t have the budget, not to mention the month it would take to get the results back during which these dogs would be stuck in the shelter.
Presenting the problem statement to the shelter’s leadership team
After going home, I decided to do some additional research and find out if this was an isolated problem. And it turns out, my local shelter is not alone at all. In fact, at many shelters, the dogs are misclassified into the wrong breeds and take longer to get adopted or face a higher rate of return back to the shelter.
One of the really amazing things about being a Cognitive Applications Technical Intern is that I get to work with some of the most advanced and cutting edge technologies that run on IBM’s Cloud. And when you work 40 hours a week on something that you think is really amazing, your problem-solving mindset doesn’t just magically switch off after leaving your normal 9-5 job. So I came up with an interesting idea: What if instead of a human “eyeballing” a dog’s breed, we had IBM Watson do it instead?
? ? ? ? ? ? ? ?The MVP running on my phone
Through a combination of open-source data sets, I was able to build a custom machine learning model that was able to determine a guess at a dog’s breed. A lot of times, one thing that really goes understated is the amount of time you need to find getting, cleaning, and optimizing the right data. I had some really great learning experiences with the data wrangling and even learned how to write some script to automate the process.
After I built the model, I made a simple iOS app for volunteers and staff at the shelter to use. The UI is pretty simple since this is all MVP: essentially you open the app, take a photo of a dog, and get the results of what breed the model thinks it is. Right now the model is trained on about 20,000 dog photos and around 120 different breeds.
With MVP in hand and research and customer discovery interviews done, I was able to go back to my local dog shelter and propose this project I built. I made a short presentation to the shelter’s leadership and they loved the idea. I’m currently in the process of making some improvements to the app and model in order to conduct a pilot program and see the actual impact this has on finding these amazing doggies a home.
This story was originally published on Joel’s personal blog at joeljoseph.space on August 2, 2019.
About the Author
Joel Joseph is a Computer Science student at the University of Southern California. This summer, he is working as a Cognitive Applications Intern at IBM in San Francisco.