ollie.leu

I have designed this carousel design for choosing your character. I also decided to implement me as a character as I work on phasing out any previous designs for a final release. This program is currently being pesky and opening new windows each time (because they are all different python programs). I don’t think this is something I can/want to fix so I will keep it like this for now. I fixed a few little bugs and worked on getting my face to show up when you speak to me instead of the previous bot’s avatar.
The memories are separated so each have different memories of your chats so they will remember different things.
Oliver’s first message is the same (working on it as I am writing this devlog) but he does have a different personality moving away from a more tutor role to more of a “I am a student just like you” kind of vibe.

I implemented a lot of stuff into the actual notes importer. It took an obscene amount of work to be able to set this up for some reason but instead of a big block of files that keep getting rewritten you can now individually remove files and then select more files and append (instead of the old setup where it would overwrite the working fileset). Funnily enough this was the bulk of what I got done. I did some more backend work fixing little bugs and commenting more code because my code is pretty confusing but overall I’m still pretty proud of the fileset selection system I implemented 😊.

Loads of optimization done to the program itself and getting it into a working state but I also adjusted the look of it to make this program a bit more ✨Girly Pop✨! I even added a cute little flower as the taskbar icon and the program window! This should be able to work with my actual speaking aspect of the program but I haven’t completely verified it (maybe it just keeps running off of old notes idk 😂). But I’m pretty proud of the visuals here and for sure will be adding images to make it look cuter and add this aesthetic to the actual AI interaction part(s) of the program!

This looks the same but I put in a lot of backend work helping the system run without doing the weird program not responding stuff and general optimizations (it actually works now lmao)!

Just a cute little interface test. I followed a general tutorial and guide for this but I’m still really proud of this one. All the buttons work (kind of) and I figured out how to do that side popup thing. Right now it has no “brain” plugged in so it’s just a dummy program but I’m having a lot of fun with the UI and I’m gonna make it all fun and cute!

I made a GitHub Repo for it! It is very messy and unorganized but you can attempt to run it! I’ll obviously update it to explain what you need, etc. but for now the aim is to have something to transfer between computers so this isn’t meant to truly be used yet.
The Repo has been removed for multiple reasons, it will get uploaded sometime soon though!

Very tiny update but I added a lot of comments so that anyone wanting to view the program can actually understand what is going on, furthermore I started work on a gui for the importer but currently have nothing to show. I implemented some ideas I had of being able to clear all imported files (with cleaning specific ones planned) because I have imported the wrong pdf or text file way too many times.

So it didn’t have proper implementation for multiple pdfs like I said it did. I really thought I implemented it but after some stress testing I learned it doesn’t do that at all so I scrambled to fix it up and it works now! It goes one pdf at a time and uses the same AI to analyze the files. I still have to add a gui but just wanted to note that I was able to fix it up! I have to go one PDF at a time because my original creation of all pdfs at once ruined formatting which is actually pretty important for my crappy local AI.
(And yes I know you can’t see the text on some of the images I post idk what to do tbh so you just get blurry text 🤣).

Not much to show but I implemented PDF to proper json. For low power AIs (like the 3.7gb gemma model I’m running) having the jsons segment the data is crucial. This (currently terminal) app takes your pdfs, extracts text (not through image sadly, just through text copy and pasting) then runs it through the same model that runs the bots (while it might not be optimal it is what I chose) and it segments it into a proper json format to help the proper bot AI run more fluidly. This will be implemented into a GUI soon.

I spent a load of time trying to get it to work and preform better, it was working with an extremely small dataset of notes and was providing low amounts of proper explanation so I put in some work and got it working nicely with more text. Boxes are still messed but auto scrolling and auto locking onto the text box have both been implemented. Whenever I try to fix the text boxes things just get worse so I’m going to put that off for now. This is gonna have to be tweaked to have it just make some rough assumptions, like in the image provided there was a lot of pointless back and forth where some assumption could have worked out much better.
My plans for tomorrow (or some other time) is to add a proper gui to enter notes into, pinkify the guis more and make them so much more cute!, and use some other character bases to give each one a distinct profile, this one will probably be my more fun “character bonding” bot whereas future bots will be more focused on proper teaching. I got some more things planned and I’m excited to see where this project goes!

I have attempted this project before with more roleplaying based bots so it used one of those as a base (as evidenced by some of it’s odd methods of messaging like actions and its dating base, the base was chosen because it is really enthusiastic and I already perfected it). Right now it is able to take notes provided through the code (the plan is to turn this into a gui format too later) and then feed it to a small local model (the model only amounts to 3.7gb in size so an extremely small one made to fit onto my 1080’s vram). This model turns it into a conversational base and asks questions. You answer and your answer is understood and either “fixed up” if you were incorrect or affirmed and you’re asked another question.
This has gone through 3 notable iterations, a start with just multiple choice questions, a roleplaying format that requires more back and forth, and this current state of the same roleplaying base but with an included gui.
There is a full gui for messaging but it is pretty lackluster as evidenced by the image. Future plans are to give it more color, auto select the text box when the program is open, and most importantly fix the messed up sizing on the message text boxes.