Building an AI improv skit generator

I find that over time I encounter a number of technologies I’m interested in trying out. Often, it’s not easy to do so in an existing project because something else is already in use that fills a similar role. A couple recent examples for me are:

• Deno, a JavaScript/TypeScript runtime and package manager
• Drizzle ORM, a TypeScript Object-Relational Mapping (ORM) tool with great performance that has syntax similar to SQL

Additionally, I learned about LangChain and LangGraph recently at CascadiaJS and was pretty excited to try my hand at building something with it. I had no need for it in my existing projects, though. Time to start a new project!

I decided to use LangGraph to build an AI agent that orchestrates an AI-powered improv show. It runs on a Deno server that exposes a single endpoint that allows users to provide a skit topic, and then generates a few rounds of banter between some AI comedians on the topic. As the agent completes steps in its task, the progress is saved as a Checkpoint which I save in a database using Drizzle.

The project isn’t deployed anywhere, but you can check out the code on my GitHub. If you want a little improv skit of your own, the repo has some setup instructions in the readme.

Learning Neovim

On Sunday I was still in the learning mood, but didn’t feel like doing any serious programming work. So instead I decided to cave to some peer pressure (and YouTube algorithm recommendations) and learn how to setup Neovim as an IDE. The experience brought back fond memories of my time as an Arch Linux user, and now I use Neovim btw. I was surprised to find that getting to a similar level of productivity with vim vs without takes very little time, and the rest is upside. Plus, editing text is substantially more fun using vim. As someone who spends most of the day typing, I believe it’s worth learning just to make the day more fun!