The Golden Apple

Welcome!
Prediction Curve
Actual Curve

Inspiration

There are more than 316 million phone users per year, and around 21% - over 66 million of them upgrade phones every year. (SellCell) With this much space for the second-hand phone market, we realized that this field can be better optimized for both the business and the customers. Introducing, The Golden Apple, your one stop shop for predicting the most accurate future about your new companions!

What it does

The Golden Apple is a very simple concept. It gives investment advice on iPhone products based on a set of events that has, and potentially might happen in the future.

It calls an agent using OpenAI API requests on all events that happened in the last projected period.
A machine learning model (Agent 2) cooks for us, mapping out all of the financial data with related historical events to create a model that can perform price prediction of devices accurate up to 97%!
Agent 3 will be taking all of that data, turning it into processable information and will be pushed to the frontend, where everyone can interact with.

How we built it

Blood, a LOT of sweat, and tears. We collected and trimmed a database on iPhone prices as well as major events (ie. product releases, software releases, major political events, etc), mapped them together and built an multi-agent model using OpenAI API and Meta's Prophet. One took care of collecting all necessary data related to events happening around the world that were Apple-based and/or other massive news, while the other processed all of the data using prices and previously-mentioned data to create a prediction model, which then ingests data into the final agent for it to provide valuable insights for the users.

Challenges we ran into

"In the midst of chaos, there is also opportunity." - Sun Tzu, The Art of War

It sure did feel like chaos. as we kept going back and forth on the idea. Even while coding the agents themselves, we still hit a lot of roadblocks, one of which, is the available dataset.

For the dataset, the amount of data points that can be taken from prices of lets say, iPhones, are not that many compared to other models. Also, most of these data have not been trimmed nor has it been put in a usable format. We had to build our own scraper tool as well as a reformatting script into JSON on our own, and that took a massive amount of time as well.

Another large problem was the learning model itself. Due to finding out very late that our model needed a lot of fixes, we had to focus a lot on that as well. Looking around and exploring choices for each agent was also a large issue that we had to come through.

Accomplishments that we're proud of

We are very proud that the project was done in time, and got relatively good results. Also, we learned a lot about communication as well as having a great vibe all around.

What we learned

Do NOT mess with time, just try our best on all ideas and making sure that they work! Also, communication is VERY important across all team members to be kept on the same page, which adds a lot to being able to finish the project in time or not. Ideas are easy to think of, but very hard to refine!

What's next for The Golden Apple

Hopefully a full on-app release so we would be able to enhance users' and businesses' satisfaction as well as being able to provide genuine support to the community, which is the most meaningful thing to us all.