Featured Project
AI Trading Arena
AI Trading Arena is Sahir Sood's ongoing full-stack market simulation platform with Claude and OpenAI trading agents, a C++ limit order book, FastAPI, SQLAlchemy, SEC/RAG evidence, WebSockets, and React.
Project Overview
AI Trading Arena is a full-stack capital markets simulation where Claude and OpenAI-powered trading bots compete against each other in a simulated stock market.
Each bot reads market prices, financial news, and optional SEC filing evidence, makes a structured trade decision, submits orders into a custom C++ limit order book, and logs its reasoning, fills, portfolio state, and PnL.
Technical Scope
- Custom C++17 central limit order book with limit orders, market orders, price-time priority, FIFO price levels, order cancellation, depth snapshots, spread and mid-price calculation, and trade logs.
- Python simulation scheduler and FastAPI backend with SQLAlchemy persistence, PostgreSQL for live data, SQLite for tests or sandbox mode, and WebSocket live event streams.
- Claude and OpenAI bot competition across BearBot, DegenBot, AnalystBot, ContrarianBot, and MacroBot personalities.
- SEC EDGAR ingestion, document chunking, optional OpenAI embeddings, RAG retrieval, evidence guardrails, and persisted evidence citations.
- React dashboard for arena summaries, bot reasoning, order book depth, trade history, sandbox controls, rankings, portfolio state, and PnL.
Roadmap
- Deterministic risk controls before every non-HOLD order.
- MCP tools for market snapshots, portfolio state, evidence retrieval, and risk checks.
- Historical replay with an as-of clock, retrieval evaluation, evidence citation scoring, and model comparison on identical replay inputs.