← Back to Work
Fetch data — historical OHLCV via Alpaca (yFinance fallback) plus technical indicators. Analyse — an LLM scores sentiment and confidence per symbol. Build strategy — allocates capital by sentiment/confidence with entry & exit rules. Backtest — simulates day-by-day with targets and stop-losses. Execute — places paper trades on Alpaca only when the backtest is positive. Evaluate — computes portfolio metrics. Self-improve — the LLM proposes refinements for the next run. Stateful, risk-aware decision pipeline with persistent memory. Provider-agnostic LLM layer and a polished CLI. Backtest gating so only validated strategies ever trade.
( Case Study · 2025 )
Trading Agent
Overview
A self-learning, utility-based trading agent built on LangGraph that observes market data, reasons with LLMs, simulates strategies, executes paper trades, and improves from feedback.
The agent loop
Stack
Python · LangGraph · multi-LLM (OpenAI / Groq / Ollama / HuggingFace) · Alpaca + yFinance