@Tim Fairley Mostly moving, setting up servers and turning them back on, and running checks on servers, networking to make sure they were running. One company we were migrating to the cloud, so I mostly helped there as well. My team mostly looked over plans, ensured security was in place like access, cold/hot storage, backups, contingency plans, etc were in place, and also ran security checks remotely. But I'm looking to step more into the construction side, possibly in a project or pre-construction role since I have some experience with implementation and rollout. I understand data movement, and also some of the protection surrounding it.

@Tim Fairley I have not heard of a role like that. I'll look into it.

@Tim Fairley Thank you! I went for the interview so I'm still awaiting feedback and response. However, I've been studying tutorials on estimating to get the jist of things. I also signed up for BlueBeam's Learning academy since that seem to have been in most job descriptions that I've come across so far.

Are you looking to just provide PDFs to the AI to read and return estimation? Or do you also want to enter or send voice commands as well?

@John Clemmer When I built other systems, the MD files were used as playbooks to help RAG reason and then to have MCP execute a tool/API based on that reasoning. I've been working with Claude more recently, and it is winning me over; however, I do wish that it would become like ChatGPT and not stop your usage if you run out of paid tokens. In my IDE, though, if I run out, I just have ChatGPT review remaining tasks, tell it the guidelines, and complete them. I'm so happy we have tools now, becaue back in the day this stuff would cost more to build out, and it takes longer. I've been able to get a decent grasp of the tools since ChatGPT dropped, and I've been building systems out of the tools ever since. A great time for ingenuity.

@Esad Omerefendic I'll say these Generative AI tools, in my opinion, are still sort of experimental, and you just have to keep security in mind to the best of your ability when building with AI tools. With RAG, you have namespaces that you will need to build least privilege access to data. I keep identity access in mind because in my field, it is just a practice. So, only a certain person, with a role, for example, can access data through a query/prompt, execute a command to the MCP to take an action. I even built a project to showcase Data Loss Prevention (DLP) on prompt and response to go beyond just the normal guardrails that are included in instructions. Ensure that you are hiding secrets correctly, because both Claude and ChatGPT sometimes code out things as MVP or Demo-like, and it can be hardcoded to expose secrets like API's, etc that should be hidden. What I like about using ChatGPT's codex is that it automatically runs a smoke test, running its checks. So I think the development teams of these AI tools know that security is an issue, but that still should be the responsibility of the builder. I use playbooks, something we do in the cybersecurity industry (or we should be doing, I won't speak for all cybersecurity), but that is also how I took the extra step to ensure that RAG would follow certain instructions as well as MCP to prevent further hallucinations and or data exposure, while my pipeline checked for prompt injection, or tool misuse, etc.

Congrats!