codeflash-internal/experiments/universe-optimize/plan.md

Top level plan:

1. Research 1000 projects on github that are important and popular and in Python, javascript/typescript and java. They are either used by important companies or are used by many people.
2. For each of the above project, start a new virtual machine on Azure with Ubuntu and 4 CPUs and 16GB RAM (choose a similar machine). On this machine download claude and setup the API key to access claude (to fill in later). When you start claude, run it in --dangerously-skip-permissions mode and with --plugin-dir=.path/to/codeflash-agent/dist. Codeflash agent should be copied to this machine from /Users/saurabh/Library/CloudStorage/Dropbox/hacks/codeflash-agent/dist. Instructions on how to use this codeflash-agent will follow
3. Fork the chosen repo on to the codeflash-ai github org and clone the fork on the new virtual machine.
4. Understand the source code of the project and try to understand what it does and how.
5. From there we will initiate optimization process start using /codeflash-optimize start command. Do read what the plugin does to know what to expect. This will setup the project on the VM, and setup benchmarks and find optimizations.
6. Codeflash agents will find optimizations and create a stacked branch with a lot of the optimizations. Collect the results of the optimizations for the next step.
7. Write an email at the end with the purpose of it being selling the results of the optimization to an engineering leader at the company of the project or the main developer of the project.