NVIDIA-AI-Blueprints/quantitative-signal-discovery-agent ? reverse-engineered prompt

Build me a Python project that can automatically discover stock market alpha signals from price and volume data using NVIDIA’s agent tools.

I want an end to end workflow where I can give it a simple request like momentum signals or mean reversion signals, and it will generate signal ideas, turn them into runnable Python code, evaluate them on S and P 500 data, and keep trying until it finds something good enough. If the signal does not meet the quality threshold, it should use the results to suggest improvements and try again. Please include a notebook version for exploring the process step by step, plus a command line way to run the same workflow from a config file.

It should be able to download fresh market data, run the evaluation, and save the best results in a structured JSON format. Add clear setup instructions for the API key and make it easy to run locally with uv. If you need to look up current docs online to match the latest NVIDIA agent toolkit usage, go ahead and do that.