Inspiration
We wanted to make stock forecasting easier because we trade stocks and it hasn't been going well for us. Most online tools are either too complex, vague, or don't have accurate results. We thought that the ideal tool for us was one we developed ourselves. What if you could see patterns from the past and use them to know what will happen in the future?
What it does and how it works
StockVision is a trading assistant that puts machine learning, predictive modeling, and real-time market data all together into one package. At its core is Chronos, a predictive engine that scans 2 years of historical patterns to find echoes, which are previous market signatures that closely match the present trajectory of a stock. For example, let’s say a stock from the past went up, then down, then back up again, and the current stock you are looking at did this as well. Then it’ll look at what ended up happening to the stock and decide based on that. This is then repeated for the rest of the data. These echoes allow us to predict the future and essentially allow you to go back in time. An important thing to note is that the data from the past that the Chronos engine uses is randomly generated based on an algorithm, but the real-time price is drawn from an FMP API. Future data will be based on this fake data because we weren't able to get an API to get that much data. StockVision even has built-in Gemini, so users can ask questions and get insights instantly. StockVision is built for both novice investors and seasoned analysts. It turns noise into clarity and doubt into action.
How we built it
We coded it using React, Tailwind, Chart.js, Financial Modeling Prep API for real-time data, and the Gemini API too. For predicting stocks, we built an engine called Chronos, which looks at simulated stock history and finds patterns. We added a Gemini Chatbot to the bottom right of the website for personalized feedback. We used some open-source Gemini to make the basis of the fake data generation algorithms, but we built Chronos.
Challenges we ran into
Working with financial APIs was very challenging. The data wouldn’t format consistently, and it took a while to find a good API to work with. One challenge we weren't able to solve was getting the real data from the past because we would have to draw on large amounts of data for it to work. It would also end up costing money if things went wrong. Another challenge was setting up the login system because it had a lot of errors.
Accomplishments that we're proud of
We are extremely proud of how well Chronos works. It finds stock echoes based on the simulated past behavior. The app itself is very well designed, looks very good, and is incredibly simple to use. It feels polished, and if we were able to use an API to get past data, we would actually use it. The Gemini ChatBot replies are also useful for people who don’t want to read charts and graphs and immediately learn what to do.
What we learned
We learned that pattern detection is incredibly hard, but making a site that can help you make money feels very, very nice. We also learned how to use AI responses better and make them a part of the app.
What's next for StockVision
We want to be able to use actual data and test how well it is able to predict future stock prices. This way, you can see how well the model could have worked in the past.
Log in or sign up for Devpost to join the conversation.