Your AI-Powered Stock Market Oracle

Harnessing social sentiment and advanced ML to guide your investment decisions.

Traditional stock analysis tools often overlook the power of social sentiment, while retail investors increasingly rely on platforms like Reddit and Twitter for market insights. We built StonkSensei to bridge this gap, combining quantitative metrics with qualitative social sentiment to democratize data-driven investing. Inspired by the GameStop saga, we wanted to create a tool that decodes "meme stock" hype while maintaining rigorous financial analysis.

StonkSensei is a next-gen stock analysis platform that:

• Analyzes sentiment across Reddit and financial news using FinBERT and FinTwitBERT ML models
• Calculates real-time "Hype Score" based on social media engagement
• Generates risk profiles using real market data + proprietary volatility algorithms
• Delivers plain-English investment theses via DeepSeek-7B LLM
• Flags blacklisted/meme stocks while suggesting whitelisted opportunities

Tech Stack

• Frontend: React/Next.js (TypeScript) + Tailwind CSS
• Backend: FastAPI (Python)
• ML Pipeline:
  • Pandas for data cleaning
  • FinBERT and FinTwitBERT for sentiment analysis
  • Custom hype/risk scoring algorithms
  • DeepSeek-7B for natural language insights
• Data Infrastructure:
  • Web scraping: Selenium + Beautiful Soup
  • Database: MongoDB Atlas
  • Auth: Okta Auth0
• Deployment:
  • AI/ML: Modal for GPU-accelerated inference

1. Model Integration Hell: Getting FinBERT/FinTwitBERT to play nicely with real-time sentiment streams
2. Reddit Rate Limits: Overcoming API restrictions with smart Selenium scraping
3. Meme Stock Detection: Separating genuine hype from bot-driven noise
4. Risk Calculation: Blending traditional beta scores with social volatility metrics
5. Prompt Engineering: Creating prompts that do proper analysis while providing data for user needs
6. Connecting Components: Making sure frontend, backend, and ML components work seamlessly with each other.

• Designed a visually appealing and accessible user interface
• Gathered large amounts of social media data using web scraping
• Utilized sentiment analysis and market data to assess risk of investments
• Used DeepSeek to provide stock advising information to users.

• Financial NLP requires domain-specific fine-tuning (generic models fail)
• Social sentiment has asymmetric impact – panic sells faster than hype buys
• User experience is king: Traders want insights, not raw data
• Web scraping ethics – when does "public data" become intrusive?
• The power of JIT model loading on Modal's GPU cluster
• Technical skills like FastAPI, web scraping, and NLP

🚀 Ready for Smarter Investing? StonkSensei – Where Wall Street Meets Main Street

In the future, we plan to utilize more financial data and improve our financial models using RAGs, LangChain, and other AI/ML systems in order to continue dominating the stock market.

This project uses the following pre-trained models:

• FinBERT: GitHub | MIT License
• FinTwitBERT: Hugging Face | Apache 2.0 License
• DeepSeek-7B: Hugging Face | Apache 2.0 License

@misc{FinTwitBERT-sentiment,
  author = {Stephan Akkerman, Tim Koornstra},
  title = {FinTwitBERT-sentiment: A Sentiment Classifier for Financial Tweets},
  year = {2023},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/StephanAkkerman/FinTwitBERT-sentiment}}
}

@misc{araci2019finbert,
  author = {Dogu Araci},
  title = {FinBERT: Financial Sentiment Analysis with Pre-trained Language Models},
  year = {2019},
  eprint = {1908.10063},
  archivePrefix = {arXiv},
  primaryClass = {cs.CL},
  doi = {10.48550/arXiv.1908.10063},
  url = {https://arxiv.org/abs/1908.10063}
}