Inspiration
Imagine losing thousands of dollars because you sold Tesla stock just hours before it skyrocketed. This isn't an hypothetical - it's the reality millions of retail investors face every day, not because they lack intelligence, but because they're overwhelmed by data. InvestAware helps remediate this problem.
The numbers are staggering: individual investors lose billions annually due to emotional trading and information overload. Almost two-thirds of retail investors underperform the market, despite having access to more information than ever before. The problem isn't a lack of data - it's having too much of it.
Picture yourself trying to make a single trading decision. You're juggling multiple news sources, technical indicators, social media sentiment, and analyst reports - all while trying to act before the opportunity disappears. This fragmented approach leads to analysis paralysis and missed opportunities.
Currently, investors face an impossible choice: either pay thousands for enterprise-level tools or settle for oversimplified apps that lack real analytical depth. Thanks to recent breakthroughs in artificial intelligence, we've found a better way.
What it does
Our platform's power comes from it's three-layered approach:
- Real-time news analysis across thousands of sources
- Historical price patterns and technical indicators
- Company fundamentals and earnings data
But here's what makes us unique - our AI doesn't just analyze these streams separately: it understands how they interact. This is best demonstrated with a real-world example:
During Apple's latest earnings announcement, our system analyzed the earnings report alongside 3,000 news articles and social media posts in real-time. The result? Our model predicted actionable insights 12 hours before a major price movement.
How we built it
We started with a two-pronged approach: one person focused on the front-end and workflow integration, while the other tackled the complexities of training the AI model. For the front-end, we began with a basic HTML/CSS/JS system, embedding necessary functions and styling directly into the HTML page using <script> tags and in-line CSS. As we progressed, we developed reusable styling sheets and scripts to ensure consistency across the website. Finally, we connected the front-end to the AI model through a Redis database and a Python back-end, creating a seamless flow of data and insights.
With the model, we faced the challenge of balancing accuracy and efficiency. Training the AI on historical data to predict news sentiment and market movements required significant computational power. To optimize this, we started by testing smaller data sets and gradually scaled up, running extensive training sessions overnight. This iterative approach allowed us to refine the model without compromising on performance.
Challenges we ran into
The primary hurdle we ran into was the lack of historical data. We tried to find a reliable resource to directly pull news headlines in recent times from, but this didn't work. This resulted in us having to create a scraper of sorts and then pull the date of the article when training our model.
Another hurdle was time. Training the model on large historical data sets to ensure accurate sentiment analysis and market predictions was incredibly time-consuming. To mitigate this, we adopted a phased approach: testing on smaller data sets during the day and running full-scale training overnight. This allowed us to make progress without being bottlenecked by the model's training time.
Lastly, we also ran into a challenge with integrating the front-end with the back-end seamlessly. Ensuring that the Redis database communicated effectively with the Python back-end and the front-end required meticulous debugging and optimization. However, overcoming these challenges made the final product more robust and reliable.
Accomplishments that we're proud of
We're incredibly proud of how the project came together. While there’s always room for polish, the core functionality is fully operational, and the predictions page, the heart of our platform, works exactly as envisioned.
Some specific achievements we’re proud of include:
- Mastering Redis: Figuring out how nodes and tasks work in Redis was a major milestone.
- UI/UX Design: Creating an intuitive and visually appealing interface that simplifies complex data for users.
- Model Training: Successfully training the AI model to analyze and predict market movements with impressive accuracy (back testing, or predicting for past dates, reveals an average accuracy of 64% ± 2).
What we learned
This project was a massive learning experience for both of us. We gained hands-on experience in UI/UX design, full-stack development, and AI model training. Understanding how to integrate these components into a cohesive system was invaluable. Additionally, we learned the importance of iterative testing and optimization, especially when working with large data sets and complex algorithms.
What's next for InvestAware
We believe InvestAware has the potential to go far beyond hackTAMS. The platform is designed with scalability in mind, and we’ve already incorporated monetization elements to prepare for a potential product launch.
Our next steps include:
- Enhancing the AI model: Expanding the data sets and refining the algorithms to improve prediction accuracy.
- Adding more features: Incorporating advanced tools like portfolio optimization and risk assessment.
- User feedback: Launching a beta version to gather user feedback and make iterative improvements.
InvestAware isn’t just a project—it’s a solution to a real-world problem.
Log in or sign up for Devpost to join the conversation.