StockGPT

Accepts free-form natural language questions about stocks, market trends, and financial metrics, then routes them through an LLM-based query parser that translates user intent into structured data requests. The system interprets colloquial financial terminology (e.g., 'Is Apple overvalued?', 'What's the tech sector doing?') and maps these to underlying market data APIs, returning conversational responses rather than raw database results.

Integrates with multiple real-time market data providers (likely Yahoo Finance, Alpha Vantage, or similar free/freemium APIs) to fetch current stock prices, volume, intraday movements, and sector performance. Implements a caching layer to reduce API call frequency and costs, with TTL-based invalidation to balance freshness against rate limits. The system normalizes data from heterogeneous sources into a unified schema before serving to the LLM context.

Takes aggregated market data and user queries, then uses an LLM (likely GPT-3.5 or similar) to generate contextual financial analysis, trend interpretation, and investment thesis summaries. The system constructs prompts that inject current market data, historical context, and financial metrics into the LLM's context window, then post-processes outputs to extract key insights. No human financial analyst reviews outputs before serving to users.

Enables users to query multiple stocks simultaneously and receive comparative metrics (valuation ratios, growth rates, sector positioning, relative performance). The system batches ticker lookups to minimize API calls, aggregates results into a unified comparison table, and uses the LLM to generate narrative comparisons (e.g., 'Stock A is cheaper than Stock B on a P/E basis but has slower growth'). Supports sector-level aggregation to identify relative strength across industries.

Maintains conversation history within a user session, allowing follow-up questions that reference previous queries without re-stating context (e.g., 'How does that compare to its 52-week high?' after asking about current price). The system stores recent queries and responses in session state, injects relevant context into subsequent LLM prompts, and manages context window size to avoid exceeding token limits. No persistent storage across sessions; history is cleared when user closes the browser.

Aggregates market data across multiple stocks within a sector to compute sector-level metrics (average P/E, median growth rate, sector momentum, relative strength vs. S&P 500). Uses LLM to interpret these aggregates and identify sector rotation patterns, leadership changes, and macroeconomic drivers. Supports hierarchical sector classification (e.g., Technology > Software > SaaS) to enable drill-down analysis.

Extracts key financial metrics (P/E ratio, dividend yield, debt-to-equity, ROE, free cash flow, earnings growth) from market data APIs and normalizes them into a consistent schema. Handles missing data gracefully (e.g., dividend yield is N/A for non-dividend stocks) and computes derived metrics (e.g., PEG ratio from P/E and growth rate). Provides both raw metrics and LLM-generated interpretations (e.g., 'P/E of 15 is below historical average, suggesting undervaluation').

Allows users to create watchlists of stocks and set price-based alerts (e.g., 'notify me if Apple drops below $150'). Stores watchlist state in browser session or optional user account, periodically polls market data APIs to check alert conditions, and triggers notifications when thresholds are breached. Supports multiple alert types (price level, percentage change, volume spike) and notification channels (in-app, email if account is linked).

Integrates with earnings calendar data (likely from free sources like Yahoo Finance or Alpha Vantage) to display upcoming earnings dates, dividend ex-dates, and corporate events. Allows users to filter stocks by event type and date range, and provides LLM-generated context about why these events matter (e.g., 'Earnings miss could trigger a 5-10% drop'). Supports setting alerts for upcoming events.