I made a fun hack at the Cloudflare AI AI Hack Night in Austin this month - took me about 40 minutes to code it up, and 5 minutes to demo it live to the audience. That included coming up with the idea, coding it, figuring out how to call an LLM model using the Cloudflare API, and then presenting it.

So what did I build? A JavaScript command line tool that fills in all of the TODO comments in your source code with code statements generated by an LLM. I used the Deepseek Coder @hf/thebloke/deepseek-coder-6.7b-base-awq model, which is a large language model trained on code.

Check out my demo here - I start at 8:01:

I really appreciated this being a non-competitive event. I used to do a lot of hackathons back in the day, and they could get competitive with the pitches. This was just for fun, and with only an hour to work on it, no one could really build anything too serious.

Some of the other presenters were Austin locals who I’ve seen at other events, and there were a couple of interns from Cloudflare showing off projects that used Cloudflare’s AI tools.

I’ve got to thank Cloudflare for putting on the event and the amazing Craig Dennis for organizing it!