FinHealth won Best Use of Streamlit out of 800+ participants at ConUHacks IX, Québec's largest hackathon.
We employed Streamlit 
to quickly build a web app that interacts with our 2 machine learning models, and to display stock prices onto a candlestick chart 📈 All built in under 24 hours! 🎉 The Devpost can be found here: https://devpost.com/software/budget-buddy-b42oin
We wanted to empower individuals to make informed financial decisions and better understand their investment strategies. Inspired by the complexity of stock data, ever-changing market news, and the challenges of personal financial planning, we built FinHealth to centralize these insights in a user-friendly platform.
- Stock Recommendations: A main dashboard showcasing stock charts, their news, recommended action (buy, sell, or hold) and sentiment (positive, neutral, or negative).
- AI Chatbot: Context-aware chat that factors in selected stock data and recent news for more informed conversations.
- Personalized Analysis: Users can submit their information and receive tailored investment advice and recommendations.
- Portfolio Analysis: An AI chatbot that analyzes screenshots of a user’s portfolio to offer deeper insights.
- Web app: Streamlit / Python
- Chatbot: OpenAI API for question answering & personal recommendations
- Sentiment Analysis: NLTK
- Buy/Sell/Hold Recommendation: scikit-learn
- Training the classification model with good results was difficult, as we attempted various permutations of features to train on (ex: moving averages). Some permutations would give accuracies above 90% - as well as training the model on multiple stocks - but the predictions didn't make sense. At the end, we found training one model for each stock gave the best predictions.
- Learning Streamlit: Adapting to Streamlit’s unique structure and deployment model was a hurdle for our team.
- AI Integration: We discovered the importance of refining prompt engineering to ensure clear and context-aware recommendations.
✅ Built a fully functional MVP in under 24 hours, with little to no sleep.
✅ Designed, implemented, and iteratively improved a classification model to give good, accurate buy/sell/hold recommendations.
✅ Built a beautiful candlestick chart UI to display stock prices, using Streamlit
✅ Won Best Use of Streamlit out of 800+ participants at ConUHacks IX!
- Integrating LLMs and Machine Learning Models into real-time applications effectively.
- Training effective and accurate machine learning models, while avoiding overfitting
- Rapid Prototyping: Streamlit allowed us to iterate quickly and incorporate user feedback on the fly.
- AI Integration: We discovered the importance of refining prompt engineering to ensure clear and context-aware recommendations.