Update 1
Almost immediately after writing my last post, I found out that one of my ideas was just launched (the same week) by someone else.
So I could still create an audio tool that does something similar for removing elements from audio, which I think is still useful and may actually produce cleaner results (akin to subtractive synthesis). That being said, this one is a bigger undertaking, and I only have 2 months where most of this GPSO work will happen exclusively on the weekends, so this idea is going in the trash.
That being said, I also have a new idea which is to steal one of Pat’s ideas and make it better. He mentioned he was working on something that would take existing customer support conversations and generate help documentation from them based on the answers provided. This got me thinking about LLM-based customer support tools, and then I started thinking about what kind of tools would actually be useful for my software business. Which led me to:
- a tool for SaaS products that takes in customer support requests and additionally takes in the codebase as context to make changes in response to the customers
- if the customer reports a bug, the code would be updated to fix the bug
- if the customer has a question how something works, the code would be read to understand what the customer is talking about
- if the customer’s message appears to be a new problem by an existing customer, we could look at recent commits to focus in on specific changes
Pat also mentioned he’s been using Cursor to build everything, which made me think a bit more about how to speed up my development for this. Like I said, I have a limited amount of hours to actually dedicate to these before the end of the year. Some of the logic I need can’t be done cleanly by LLMs, but it is worthwhile delegating more tasks to my coding LLM tool (Supermaven).