Inspiration
We were inspired by the increasing number of young investors entering the market who lack personalized guidance when building their portfolios. Most tools focus on past trends or generic recommendations, but we wanted to create something smarter—something that understands the investor as a person, not just a wallet. What it does
Our platform takes a natural language investor profile or structured data input (like employment status, age, dislikes, salary, etc.) and generates a customized stock portfolio. It uses LSTM-based price predictions, volatility analysis, and a risk-informed asset allocation strategy to recommend an optimal, diversified, and personalized investment basket—all within the user’s budget and preferences. How we built it
We built the system by integrating several components: • A profile parser that converts raw user input into structured data. • A rule-based system to assess investor risk. • A portfolio builder that filters stocks based on volatility, sector preferences, and score-to-volatility ratios. • An LSTM model trained on historical stock data to predict future prices and assign a score to each stock. • A dynamic allocator that calculates how many shares of each stock to buy within budget.
We used Python, PyTorch, Pandas, and Matplotlib for the backend, with Jupyter for rapid prototyping.
Challenges we ran into • Handling sparse or missing data in the investor profile. • Making sure our LSTM predictions generalize well across volatile markets. • Balancing diversity with optimization—too much filtering made portfolios too thin. • Aligning budget constraints with price rounding when buying fractional quantities isn’t allowed.
Accomplishments that we’re proud of • Built a fully functioning, profile-based portfolio builder from scratch. • Integrated deep learning into practical finance decision-making. • Created a modular system that could work for both beginner and experienced investors. • Designed it to respect ethical investing (e.g., excluding unwanted sectors).
What we learned • Combining machine learning with rule-based logic gives the best of both worlds. • User context matters just as much as market data in financial decisions. • Real-world budgeting (like price rounding) introduces tricky constraints. • Simplicity in model interpretability is crucial when decisions involve money.
What’s next for Portfolio and Web Scrapers • Add support for ESG filtering and responsible investing. • Improve the LSTM model with sentiment data from financial news. • Develop a real-time dashboard to track portfolio value over time. • Enable voice-based investor profile input and chat-based feedback. • Explore integration with actual trading platforms for seamless execution.
Log in or sign up for Devpost to join the conversation.