The author developed a series of AI agents for personal use, ranging from simple text analysis to complex multi-agent systems. The first agent performs article summarization, identifies logical fallacies, and provides arguments for and against the text's thesis. A second agent manages personal health data by processing medical PDFs, tracking health metrics via Xiaomi smart devices, and building a health history.
For financial decision-making, the author created a multi-agent system to forecast currency exchange rates. The system uses adversarial pairs of agents—an optimist and a pessimist—to analyze fundamental and technical factors, with a judge agent providing the final forecast. This approach mirrors the AlphaAgents research published by BlackRock in August 2025, which uses specialized agents to manage stock portfolios.
The most advanced implementation is a pipeline for quantitative investment strategies. It functions as an automated office where agents handle research, code development, review, and risk analysis. The author maintains human control over the process, approving strategies before they go live. This system has significantly increased efficiency, allowing the deployment of over 30 strategies in one week, compared to the previous timeline of several quarters.
The infrastructure relies on asynchronous communication tools like ntfy for notifications and web interfaces for remote interaction. Reports are managed via Google Drive, and a shared knowledge base allows agents to learn from past iterations. The author emphasizes that clear specifications are critical for agent performance, a concept referred to as Spec-Driven Development.
