AI in finance is powerful but requires careful consideration of market efficiency and regulatory constraints. News sentiment analysis: NLP models trained on financial news (FinBERT is BERT fine-tuned on financial domain) extract sentiment for specific tickers. Aggregate sentiment score correlates with short-term price movements but effect is eroded as more participants use similar approaches. Earnings call analysis: transcribe earnings calls with Whisper, extract management sentiment, forward guidance language, question evasion patterns. LLM summarizes key risks and opportunities from calls. Alternative data: satellite imagery of retail parking lots, shipping container counts, mobile foot traffic, social media mention trends. Signals are valuable but fade as more funds use them. Technical pattern recognition: LSTM models trained on OHLCV data predict pattern continuations. Performance varies significantly across market conditions. Best used as one factor among many. Portfolio optimization: Black-Litterman model combines market equilibrium with your views (AI-generated or analyst). Mean-variance optimization with robust covariance estimation. ML for factor exposure optimization. Risk management: AI for VaR estimation with non-normal distributions, stress testing with ML scenarios, real-time margin monitoring. Regulations: FINRA Rule 3110 requires supervision of algorithmic trading. SEC regulations on automated investment advice. Important disclaimer: past performance does not predict future results. AI-generated investment signals are not investment advice.