Should I Buy Gas Today or Tomorrow? (Pump Predict~)

Pump Predict monitors global supply chain news and crude oil term structures to tell you exactly when you should buy gas to save money~ View the project here: https://pumppredict.streamlit.app/

Comment

Inspiration

As a commuter student, I normally drive about an two hours to and from campus. Given the current economic situation, the price of gasoline has increased dramatically. As someone who has some background in finance, math, and computer science I thought it would be useful to create a model to help commuters based in the United States save money when fueling up. The Pump Predict project is my solution :)

What it does

Pump Predict is a Vector Autoregression Model (VAR) which uses 4 forms of input - the spot price of WTI Crude oil, wholesale price of gasoline, consumer price of gasoline, and a sentiment analysis of oil-related news. The VAR model uses these and predicts the amount in cents that the price of oil will move towards in the next 5 days.

It will simply answer "Should I Buy Gas Today or Tomorrow?"

How we built it

I primarily used Python for the data analysis (Pandas, Numpy), the frontend (Streamlit), and the backend code. This project was aided by the use of several LLMs, including Copilot, Gemini, Claude - also my lack of sleep :)

Challenges we ran into

This was the first project where I primarily used AI to code. I had to learn how to use several AI tools, develop continuously on GitHub (commit, branch, and conflict management), but most importantly debug A LOT. Most surprisingly, a large challenge was to come up with the design of the tool, including the concept, UI, and backend system design.

While the Pump Predict has (somewhat) useful insight, it is severely limited by the APIs it depends on. Additionally, the accuracy of the model is questionable. With more time the model could be improved for a more reliable tool.

Accomplishments that we're proud of

  • The data is dynamically gathered through multiple APIs
  • Customer-focused user interface
  • VAR model works, to a limited extent
  • Use of AI to speed up development and g
  • Project is 100% Python
  • THIS WAS FUN!!!

What we learned

  • Expect the unexpected - simple items became difficult to implement when considering the app as a whole
  • Using AI isn't just about prompting it. You must rely on your intuition to do good system design and come up with a great idea.
  • Modeling is difficult, but rewarding
  • The fun part wasn't just in building something useful, but learning alongside friends and gaining skills I will apply to my future projects :)

What's next for Pump Predict?

  • Finish the Local Gas Finder feature
  • Improve the VAR model
  • Track/rate which gas stations have historic low gas prices
  • Determine the most convenient/cheapest gas stations of where one should actually fill up gas on a trip
  • No current plans to continue development of the project, unless there is demand for it :)

Built With

Share this project:

Updates