C.A.R.T.

Inspiration

Hackathons are always interesting to see how talented groups create diverse and complex solutions to the same prompt. We all realize that it's hard to grow as a developer without leaving your comfort zone, so working in AWS and attempting to fully productionize an idea from start to finish in about 2 weeks (while balancing full time jobs) certainly seemed like a challenge.

What it does

CART allows staff to get a 360 view of the customer. The Data Lakehouse enables transformation on data that will come from multiple sources, ultimately being fed into multiple advanced ML/Gen AI functions. These results are sent to a web application where staff can easily digest the most important data for making each call a success. Features include customer LTV, forecasting, product recommendations, email generation and more.

How we built it

CART relies on the Data Lakehouse to ingest and transform data. Databricks allows a single stop shop for all things analytics, ML and Gen AI. We used a variety of Data Science libraries and functions to create the models. We used several integrations of Databricks APIs to feed data to the app. We built the entire front end in Streamlit, which provided an easy to implement the APIs.

Challenges we ran into

We all took a step outside our usual code stack by learning AWS and Streamlit. There are always bumps when learning a new technology.

Managing cost is always a priority when doing personal projects. Databricks makes that easy with all of the custom cluster sizes and reporting.

Accomplishments that we're proud of

We are proud to present a functioning prototype of something that can provide real business value to a company. There are no static codes or patches, all of the data and functions come straight from the adventureworks database - something we had very little experience with. Putting all of this together in 2 weeks, while also balancing full time jobs was certainly something to be proud of.

What we learned

We each used this opportunity to learn a new tool or library in order to complete this project. This was our first time utilizing Streamlit in a pseudo-production environment, and Vector Search was almost entirely unknown. Having experience in Azure, setting up an AWS environment from the ground up was also beneficial to see. Lastly, GenAI is likely going to continue being a hot topic moving forward, so the exposure is both exciting and valuable experience.

What's next for C.A.R.T.

CART can be used for a POC of a production app for customer service representatives. One of our teammates thinks his company will be very excited to see the project.