Deep Learning & LLM Trading Strategies

A Workbench, Not a Single Bot

This repo holds several parallel attempts at the same problem, predicting price direction and generating trade signals for forex and crypto, built at different times with different approaches. It's a set of prototypes to mine ideas from, not a working bot pointed at a live account: models are not currently profitable against the thresholds documented in the repo's own criteria notes.

Two Different Approaches

• LSTM classifiers (the Simplified Trading bot and Trading_bot_copy folders): predict price direction (up/down) rather than raw price, trained on engineered technical-indicator features, RSI, MACD, ATR, Bollinger Bands, OBV, CMF, KST, plus a separate news-sentiment scraper.
• An LLM agent (Tradegpt): skips training entirely. It prompts an LLM (OpenAI, Gemini, or DeepSeek) with recent price data, a hand-written RSI + EMA + sentiment strategy, and asks it to reason about the trade, then backtests the resulting signal with matplotlib.

Pipeline Shape

Each sub-project shares the same rough pipeline: pull OHLCV market data (yfinance / Alpha Vantage / Twelve Data), engineer technical-indicator features and classification targets, score recent news sentiment, then either train an LSTM classifier or feed everything into an LLM prompt.

What's Next

The repo's own notes track the open problems: removing sentiment from the trained LSTM features (it may work better as a post-hoc adjustment, as in the GPT approach), retraining periodically rather than once, and merging the three near-duplicate LSTM pipelines into one shared package instead of copy-pasted experiments.

Tech Stack: Python, TensorFlow/Keras (LSTM), yfinance/Alpha Vantage/Twelve Data, Selenium news scraping, OpenAI/Gemini/DeepSeek APIs

Educational project, not financial advice. Nothing here is currently profitable or production-ready.