RD-Agent local setup and finance agent review

Uploaded on March 18, 2025, this tutorial covers Microsoft's RD-Agent, an open source platform for automating research and development. The source names no RD-Agent release version or separate recording date. Its setup details and limitations describe the demonstrated implementation at that time.

The speaker clones the repository, creates a Conda environment with Python 3.10, installs RD-Agent through pip, and configures an environment file with an OpenAI API key. He also mentions Python 3.11 as an option and Azure OpenAI as another backend. Docker is required for the finance demonstration. Although the agent runs on the host machine, this setup uses a paid hosted model. The speaker says local models are unsupported in the implementation he demonstrates; this is not an offline inference tutorial.

The finance AI agent generates hypotheses, defines financial factors, writes implementation code, and performs quantitative validation and backtesting. The speaker explains these steps through the execution logs, but the demonstration does not establish trading accuracy or profitability.

His review focuses on practical limits: lengthy runs, cumbersome customization, and documentation he finds inadequate for building an agent for another domain. He reports that a previous finance run took about 30 minutes and cost roughly US$15 in API usage. That is his reported example, not a fixed price or a general estimate for other workloads.