← Back to Blog

Shaco

Shaco is my first iteration in creating my own AI assistant, completely offline.

Shaco, an anagram of chaos, is my idea that something cool can be created from chaotic information and inputs. Combining my life into something that can be worked with from a second take. I want to create logical reasoning which focuses on the truth of my situation without dodgy political biases etc.

It is somewhat benched until i grasp more about coding fundamentals. So for now, I will describe the concept behind it and why I want my own AI asisstant.

The Why

It is not my intent to compete with huge datasets LLMs are trained upon. This assistant is to asisst my everyday life.

Imagine;

  • A system that can control your house, lets you know if youve locked the house. small ‘fun things’
  • reminders, plans you have that day.
  • has an understanding of your vitals, exercise history.
  • a second brain that can analyse finances and youtube videos. a second take on what I should invest or talk about online.

Now I suspect the above might sound scary. but I could provide it with logical reasoning that is truth based, not politically orientated. its interests are for me, sure, but I would not want harm for others.

This is the problem with public LLMs. they have biases based on those who develop them.

The How

The reason this is on pause because Im not sure I have gone about this the most effective way. For instance;

Im using a common text-gen-ui wrapper for the model mistral 7B, connected to various modules it can call upon. I will also create some form of RAG setup it can reference, using data such as my youtube videos.

Problems so far:

  • I have an AMD GPU.
    • This means I have to run this on my CPU. I can’t fine tune it either.
  • my lack of understanding
    • I vibe coded my way to getting this far. this is why I have started from the fundamentals of learning python.
  • There are other ways to building your own LLMs
    • I will want to create a shopping list of frameworks or systems that can give me greater control on building my own AI assistant. Will definitely look into different NLP technologies too.

Next steps:

  • Keep whatever data is currently in the project
  • Refactor the modules and shaco_core. make them useful, code them the hard way.
  • reconfigure the LLM setup entirely.
  • invest in a new PC
    • or, as im not sure where I will be living anytime soon, upload it onto some virtual platform for now. It’s all on github anyway (just not public)

Note: this project is hidden on my github for now.