Incorporating LLMs into software is a software dev job from what I have experienced. You don't need to know anything other than how to design and integrate APIs. So, the real answer to your question is dependent on what companies are actually "building" something versus which are simply slapping together a bunch of things (I am not against either, if my phrasing suggested otherwise!). A lot of companies and research groups are emerging where you do need ML theory and hence ML engineers. Take for example PPFL (Privacy Preserving Federated Learning) or similar fields.
Pretty cool! Are you scraping the data first to pass it to an AI or letting it do the entire thing by itself? Also, if its the latter, are you using a multi-agent framework? Only asking because it seems way too detailed for a single input-output sequence!
Working on an automated OSINT tool. OSINT is a tough nut to crack because there is so much one can do and the web is literally infinite. So, this tool can automate browsers, it can call functions in a ReAct loop and make inferences on that. It essentially a combination of an investigator + a tool.