I Built a Stock Price Prediction App While Learning ML - Here's What I Learned

TL;DR: I built a full-stack stock prediction app with 7 ML algorithms, 1,001 stocks, and offline-first support. It's open source, and I'd love your feedback. 🔗 github.com/ankit02327/stock-price How It Started This project began from something I was personally learning: machine learning. While taking ML courses at university, I wanted to understand how the algorithms I was studying could be combined into a complete, practical application. That led me to build Stock Price Prediction around a real-world problem instead of keeping the work inside individual coursework exercises. The educational purpose has remained central. I want someone learning machine learning to be able to look at a real application, understand how different models are used, work with real financial data, experiment with the code, and eventually build projects of their own. What It Actually Does - 7 ML algorithms: Linear Regression, Decision Tree, Random Forest, SVM (basic models) + KNN, ARIMA, Autoencoder (advanced models) - 1,001 stocks: 501 US + 500 Indian (936 used for training after filtering insufficient data) - Real-time data: US stocks via Finnhub, Indian stocks via Upstox, with permanent offline storage fallback - 5-year historical analysis: Interactive charts with Recharts - 38 technical indicators calculated from historical data - Currency conversion: Automatic USD/INR conversion - Smart training: Percentage-based predictions with confidence scoring The Tech Stack | Layer | Technology | |---|---| | Backend | Flask 2.3.3, Python 3.8+, TensorFlow 2.20, scikit-learn, statsmodels | | Frontend | React 18, TypeScript, Vite, Tailwind CSS, Recharts | | APIs | Finnhub (US), Upstox (India), yfinance (historical) | What Makes It Different: Offline-First The system works completely offline without any API keys. This was a deliberate design choice - I wanted anyone to be able to clone the repo and start experimenting immediately, without signing up for API keys or worrying about rate limits. What works offline: - Stock info for 1,001 stocks - Complete 5-year historical charts (2020-2024) - All trained ML models - Full-text search - 38 technical indicators What needs API keys: Live prices only. What I Learned Machine learning is not magic. It finds patterns, but you need domain knowledge to make sense of them. Data quality is everything. Stock data is messy - splits, dividends, missing values. A robust preprocessing pipeline matters more than a fancy model. Start simple, then iterate. My first model was basic linear regression. It wasn't great, but it worked, and it gave me a foundation to build on. Open source is a two-way street. The project now has 27 contributors and 14 releases. I've learned as much from the community as from building it. What's Next I'm planning substantial improvements: FastAPI migration, Redis caching, PostgreSQL, Docker Compose, MLflow experiment tracking, news sentiment analysis, and a backtesting system. The project is still growing. I see its current US and Indian market coverage as a starting point rather than the final scope. How You Can Help If this sounds interesting: - ⭐ Star the repo: github.com/ankit02327/stock-price - 🐛 Open an issue if you find a bug - 🔧 Contribute - there are "good first issue" labels for newcomers - 📣 Share it with anyone learning ML Let's Connect - GitHub: @ankit02327 Thanks for reading. If you're learning ML too, I'd love to hear what you're building. Top comments (0)