Working on the flex attributes project for invoices and quotes inside the mini scan cart. Reviewing code for yesterday and then jumping back in on the project. Small work session.

These are just some ideas to help with the AI agent training that we need to do.

- Use ChatGPT to help with the training

- Ask it questions about what might be missing - ask it to ask me about that? (Helping us to find what might be missing)

- Use it to train other AI engines and models

- I'm thinking about a general layout and organization structure (for a master adilas document) similar to the adilas presentation gallery. Help that go even deeper and deeper to show the different parts and pieces. That would include business functions, application player groups, desired outcomes (coming - currently called attributes but we want to redo that section and rename it to desired outcomes), and core concepts.

- See what it can pick up by crawling the site. Then help to feed it specific information and help put things into context and additional details.

- Take your time, not too much info at once, and then pause to make sure that we are good to go. The pause is important.

- I would love to do some of the training in levels... intro, beginner, intermediate, advanced, and deep or expert level training. That would be really cool!

- We are on purpose a general business tool, help the AI agents know and understand that.

- Fill in the gaps, let it do what it know how to do.

- On help files (existing files), re-read them and make sure that they can be used as standalone pages. Add a small mini blurb or description at the top. Don't assume that people know what each help file is for... They may get to it from a different source vs naturally navigating to that page based on normal page flow or business processes.

Recording notes from 11/24/25 to 11/27/25.

Work session with Bryan Dayton on flex attributes for invoices and quotes. We are getting closer. Working on showing both invoice flex attributes and quote flex attributes inside the cart. Lots of moving pieces. Bryan was fixing some merge conflicts and I was coding the new logic for showing and updating the flex attributes. We got all the way to inserting the new flex attribute records. We ran into a little snag but will hit it again with fresh minds and eyes. Good long work session today.

Touching base with Abby. She has been working on some vector art training and is doing great. We jumped in and looked around on her practice project and talked about a bunch of things. That was fun. She is catching on very well. We also talked about other training in the photo editing and desktop publishing side of graphics and digital art. We then got into a number of other topics and talking about optimizing images, image sizes, dpi, resolution, and other semi technical pieces. Towards the end, we were even talking about saving your practice documents. We revisited some original ideas on the adilas logo and where that inspiration came from. See element of time # 4144 in the shop for some graphics of where the adilas logo ideas came from.

Meeting with the Nxtlinq guys. Talking about advancements on persistent memory and where things are going. We talked about the backend Nxtlinq dashboard and how to read and help the agent really learn (make things stick). We may need to focus on a specific part of the application and get it really tight. Due to Thanksgiving, we will meet up again in a week. In the meantime, I will be working on some other projects. Once ready, we will get back to AI agent training.

Quick touch base with the AI agent to see if it was able to remember anything from the training session the day before. Sadly, still no long term data retention. Recording notes and formatting the stuff from the AI training session.

Got a text from Steve and we jumped on to the gotomeeting session. We were talking about the AI agent and where we are at. Steve had me check ChatGPT for something that he put in there just the other day. It was able to find it. We are thinking that we may need to train up the main or mother AI model or AI engine. We may need to look deeper on that. I also went over what Shannon and I were talking about this morning (see element of time # 12458). Good conversation.

Working with Shannon. Talking about the time it takes to use the AI agents. Talking with Shannon about the type of AI tool we are trying to use and setup. We spent some time talking about concerns and possible constraints, moving forward.

My constraints:

- I'm willing but should we do this? I really want it to work - I can see the potential - We need a master trainer that can help others

- We need to communicate - Steve and I

- My time

- Already put in a ton of time

- Questions on funding and budgets

- Train the mother AI engines and models or the mini model? Which one(s)?

- I need some direction - too many things on the plate and I don't know what to prioritize

- The training pieces are all over (on the web and in certain docs) also in my head - No one spot for all training currently

- Things keep changing

- I am saving my conversations with the AI agent, but it takes time to format them, and if it (the training) doesn't stick, then what?

- Putting in energy but not getting the output that is desired - we can only go fast for so long - we can keep chipping away at things, but that is it

- There are lots of parties involved, adilas, Nxtlinq, our users, and the actual AI agent

- We have a vision and we can build a number of things out... we have plans - those plans will take months and months - kinda feeling pressured

- Do we really want to have meetings every two days

- Tons of potential... how far are we willing to push it?

AI agent constraints:

- I've done x hours on the mini model, currently not retaining anything

- Storage at the database level - For them and for us - How can we help this process?

- How can we help with storage - Known needs - System level, corp level, and user level

- They are retaining the chat messages - At the Nxtlinq level - We are not

- Question - How do we really train this agent or entity?

- It, the agent, is giving bogus answers - For example - How do I get rid of an element of time - How can I tell how many elements of time I have (see stored chats from 11/25/25 - Some other user, not from me)

- Can it do what we need it to do

- Cut off from the mother, meaning the agent or mini model - Already over a year behind the data on the mother AI engine

- We don't know the model that it is based on... It really likes the security and blockchain stuff - Meaning what was the original purpose of the AI agent.

- We just need a simple bot that can translate natural language into adilas tools and prompts

- Not retaining information long term... Currently only per person, per session, and that's it

- How are they doing with the crawler to pick up content

- Demands - too much on us, build, train, market, deploy, and it doesn't even work all the way yet

- People aren't wanting it... what it currently is and can do... It take too long, it requires too many two-part authentication stuff, and it can't do advanced or multi step things...

- It is being used less and less - People want it, but if it gives them the wrong answer and isn't fast enough, they will move on

- Too slow, not helpful enough (wrong answers), too many auth steps, can't do complex actions, etc.

- Say you aren't using it... then you decide to use it... and it takes multiple steps to do the auth stuff... you have a quick question and you want an answer in a couple of seconds... you then have to do a multi-step 2-5 minute process to get your answer. Pretty soon, you say, no, not doing that... that is from our users.

At the end of the session with Shannon, we were talking about our simplified AI plan - 1. Teach it how to do some basic or system navigation, 2. Get it all trained up so that it can help with training and support, 3. The ideal is that it knows everything and can help with the higher consulting level and business level questions.

Shannon and I were also talking about the jellyfish model and helping to stabilize the team and their roles. There is a need there.

Doing some AI agent training inside of adilas. Training on help files and adilas elevator pitch. More AI testing and even doing some testing and training from different servers and different user id's. See attached for some of our notes.

Emails and recording notes from last week on 11/20/25. Trying to use the AI agent a little bit more. Currently, it is kinda slow... "Thinking...". I can do it way faster using the AI quick search or the plain adilas quick search. We'll get it all trained up.

Emails and logging in and checking out the Nxtlinq backend dashboard. I'm hoping that I can use that or the main chat interface to help train the AI agent. Either way, we are making progress. We'll see where it goes.

Quick testing of the AI agent. It is still not retaining all of the information that we are training it on. We have to figure out a way for it to retain the correct background information (persistent memory). see attached.