StockTrAIder: An Open-Source Platform for LLM-Driven Stock Market Insights via Curated ML Context

Existing tools are paywalled, use proprietary data, and are not accessible to students, small businesses, or casual investors. This work presents StockTrAIder, a free and open-source AI/ML-driven platform that aims to bridge this gap by building custom, curated contexts for LLMs to generate real-world stock market insights and predictions. 

Freely available external APIs, custom ML models (trained on freely available data), and a list of the most recent news articles covering real-world political events are provided to the LLM to synthesize real-world insights, such as how geopolitical events may impact the stock market. gpt-4o-mini is used to produce weekly market summaries, stock picks, and buy/sell signals. The platform is a full-stack web application (React and FastAPI) deployed via GitLab CI/CD, with front-end and back-end containers built with Docker. 

Extensive validation experiments were evaluated on the AI and ML models. The AI (LLM) models were evaluated over 10 weeks using five metrics, with five trials per week. All models achieved nearly identical performance, with the selected model achieving 99.7% symbol accuracy, 27.3% symbol coverage, 100% hedge language compliance, 37.4% faithfulness, and 24.5% relevance. These metrics reflect a balance between reporting data and providing analysis on how that data will impact a stock or the market as a whole. The ML model, XGBoostInvestor, is a bagged model of XGBoostRegressors that is backtested from 2019 to 2026 and outperforms the S&P 500 in six of the eight years, whilst minimizing major losses when the overall market is bearish.

StockTrAIder is a comprehensive platform of carefully designed LLM insights that are driven by ML models, historical data, and select external API context, which enables users to look at a high-level overview of market conditions, or dive deep into the details of a stock, enabling learning and profitability to the users, all within free API usage limits and free to users.

Committee: Simanta Mitra (major professor)