@kaitlyn

Added Google Workspace MCP

Ran into a few issues with this one. First the server is hosted locally but wasn’t packaged so i had to first clone the open source repo. This meant I also had to declare the current working directory. Another issue was that some parts of the Google Workspace schema didn’t sync well with Groq. So i had to clean the function names and I had to get rid of an attribute called pattern. Another Issue I had was that when I asked for the list of tools, my function was returning a pydantic list rather than a typical one.

But now it can access my Google Workspace and can make changes when I prompt it.

Next steps are to clean up the pipeline so there is a clear flow of events using a small routing model to pick either the LLM or the tool calling.

I began working on E.V. an AI assistant just like JARVIS from Ironman. However, the namesake is from the new Spiderman movie which inspired me to finally take this project on.

So far I have:

1. Implemented a simple chatbot using the Groq api
2. Added Speech to text and text to speech
3. Added in tool calling features
4. Made a script to pull tools from a locally hosted MCP
5. Added a feature to call tools from locally hosted MCPs

Working on not only recommending books but also using ML to recommend genres.

ran into some issues with the dataset I am using being too sparse so I tried some work arounds to use clustering to make the genres less sparse. It is still a work in progress, but I have added genre groupings using Kmeans and cosine similarity from averaged vectors. Both are still running into some issues stemming from the original dataset being quite sparse.

The recommendation system now recommends books based on a list of favorite books. Additionally, I began adding a clustering feature for genres, but this has not been finished yet.

Finished the schematic for the macropad pcb and began laying out the pieces. Changed 3x3 switch matrix to make one of the switches in the matrix for the rotary encoder button.

Worked on macropad from the guides.

For some reason the first 2 hours of recording didn’t go through.

But then I decided to upgrade my macropad and I am working on the PCB Schematic. I want 9 keys, a rotary encoder, LEDs, and an OLED screen. Decided to make a 3x3 switch matrix.

I began work on my Book Recommendation System.

First found a dataset off of Kaggle with a comprehensive collection of information on each book. I wanted to do content based filtering so I made sure the dataset had descriptions and genres for each book.

First, I cleaned up the data, created a “combined features” column that took all the string data and combined it. Any book that didn’t have the string data was disregarded.

Then I used TFIDF Vectors and Cosine Similarity to find relating books.

The similarity matrix was also visualized with Matplotlib.