Midas — LLM-generated trading strategies, backtested
A trading research platform where the LLM writes the code: strategies are generated as C# by a model via OpenRouter, executed against historical data in a backtesting module, and the results stream back to a React dashboard over SignalR.
A companion Semantic Kernel service exposes trading-analysis tools through both REST and an MCP server, and a dedicated scraper ingests market data. The core idea: treat strategy ideation as a generate-execute-evaluate loop instead of hand-coding every hypothesis.
Project information
- Technology used: .NET 8 modular monolith, React, PostgreSQL, SignalR, Hangfire, Semantic Kernel, MCP, OpenRouter
- Project goal: Close the loop between LLM strategy ideas and measurable backtests
- Date: 2025