Data Science, Pasta Sauce & a Spring of learning!

It has been a busy few months at Walmart’s Emerging Technology office in Plano with business deliverables, production rollouts, production support of existing systems, hacks/experiments on diverse areas ranging from CV, ML/DS, IoT etc.…. Amidst all this awesome fun, in January of 2019, we decided to capitalize on a great opportunity to continue our partnership with academia in TX and embarked on a 4-month Spring capstone project in collaboration with UT Dallas. We had great backing from our Senior Directors (both our Dallas site lead, Andy Minteer and IoT platform lead, Carlos Riojas) to execute a project that involved IoT and could realize future business value.  After a couple of discussions, we decided that it would be worthwhile to focus our scope more on the Data science side of IoT and decided to enlist the support of one of our Data Scientists, Dr. Neelabh Pant to lead the project. Neelabh and I had a great 4 months working on this effort along with our diligent crew from UT Dallas and this is a humble attempt to highlight some of the great fun and learnings we encountered!

You might wonder if our project was related to Pasta Sauce and Data Science after having read the title ?? Keep wondering on….. It will all make sense when you finish reading this… ??

Too many brilliant ideas, which one to pick?

We did initial prep work in the month of December (2018) vetting through the feasibility of various ideas on the project’s scope. Our tech product owner for IoT, Joshua Segraves and analytics team members that were working closely with our business partners in the home office had several ideas, and we spent considerable time with these stakeholders to hear everyone’s thoughts. Neelabh was a huge advocate of picking scope that could challenge the students and generates learnings that can be applied to existing systems. We decided that providing the students a massive data set (at Walmart’s scale) might not be the most productive usage of time and decided to instead carve out a narrower dataset focused on a very specific business problem.  We met with our business partners that have responsibilities for Walmart’s Energy needs and got their blessing on the scope of the project. Our final idea was to execute a project to understand Walmart’s existing energy costs, forecast future energy costs using a variety of ML techniques and to visualize the results innovatively. Dr. Pant wanted to focus on helping the students understand the basics of ML and DS from ground up and wanted to focus on teaching them a variety of modeling techniques, algorithms and tools that they could each use for their future careers in tech. All set, we were ready to rock and roll with our project proposals by mid Jan!!

Hello, Team! Let’s get this show started…

Project proposals were put in front of the students in the final week of January and a team of students picked our project! We were incredibly excited to begin the effort. We had specifically asked for students that were competent in 1 or more programming languages and understood ML techniques well.  Our team consisted of 4 Under Graduate students and had a senior student as mentor. We kicked off the project meeting with the students, and other key stakeholders. We set the cadence for meeting the team in person once every week at UTD’s campus and staying in touch on a regular basis using GroupMe. GroupMe turned out to be a great idea because it allowed quick communication and allowed Neelabh to respond quickly to technical questions.

Blood, Sweat, Tears and hey, hey, hey, Pasta Sauce ??

Our team of 4 consisted of Srinath Shankar, Sriram Bhat, Steve Chen and Siddheshvar Kannan. They were  super smart and hardworking with a great sense of humor. We had a great time getting to know each of them as individuals. Srinath was confident and calm and had a future job awaiting him at a banking firm. Sriram was the group’s front-end developer with an affinity to React. Steve was the quiet one, the silent listener who always had great insights. Sid was the Walmart Product SME (Subject Matter expert) in the group – he had great knowledge of Walmart’s products and especially Pasta Sauce as he constantly roamed Walmart’s aisles in search of best priced deals. We had several conversations about Walmart’s pricing strategy, core competencies and how Walmart’s everyday low cost choices helped students navigate hunger pangs at odd hours. Amidst these fun conversations, the project work continued to progress smoothly. The students downloaded the tools we asked them to, began analyzing the business domain and the dataset and started joining weekly meetings with stakeholders to clarify various aspects of the dataset. Neelabh consistently challenged them to dig deeper and try new methods and pointed them in the right direction if they went off the course. Neelabh and I also had a great time driving to UTD every week (I did slack off on these drives towards the later half of the project claiming higher priorities, but he kept at it!). Our drives were full of fun conversations on various topics ranging from cricket, politics, family, work to Daler Mehendi’s songs ??. 

Towards the middle of our project timeline, we were able to host the students at our office in Plano and they loved the tour of the office, our Foosball tables and interacting with our broader team of data engineers, data scientists and developers. We loved showing off some of our innovative projects to them and they enthralled us with curious questions.  For me, the most memorable part of the Capstone experience was the camaraderie that we forged together as a team. 

Progress made…

In the real world, data scientists typically work with data engineers who handle the efforts of retrieving data from various source systems, cleaning and organizing the data to enable the creation of ML models. For the Capstone, we intentionally decided to minimize the data engineering efforts in order to focus more on data science. Hence, we provided the students a highly sanitized dataset to use. Over the course of four months, the team created multiple models to forecast energy usage and continued to engineer features on their models to increase accuracy. They dabbled in visualizing the models through a website. The team learnt various data exploration techniques such as correlation analysis, distribution analysis etc. They validated their models using testing datasets and evaluated the pros and cons of each model. Their final presentation was a summary of all the models created, the advantages and disadvantages of each model and their final recommended choice for the most accurate model. We were present in person at the CS UTD Expo 2019 to hear our team present their work and I must admit, it was a very proud moment for all of us.

It’s all about that jazz…..

Amidst all our focus to ensure the students maximized their data science learnings’, we did not realize that all of the Capstone projects are also pitted against each other for a final competition that actually anoints winners.  Unfortunately, our team did not win the competition even though they delivered great results and presented with poise and confidence. We realized as we saw the winners, that we could have jazzed up our solution a few notches higher by combining elements such as AR/VR, CV, etc.…This was a good learning for us and we felt good about the fact that our team had worked on something that added business value but at the same time also had scope for future improvement. Hey, there is always 2020’s Capstone where we can kick it up a few notches higher ??. On the final day of the CS Expo, our students also honored us for our support by gifting us…… yes, yes, yes, you guessed it right, Pasta sauces from Walmart!! ??. They also had written a very sweet card for their Tech lead, Neelabh thanking him for his patience and leadership. Aaahh…. A sweet moment indeed. It was a successful culmination of 4 months of effort and I am incredibly thankful to all of the partners that made this possible. I look forward to another season of learning with a new team.