Tarix

The first try with embdedding on my CPU

Inspiration

My inspiration for this project stemmed from me learning about the different tariffs that have been placed on foreign countries over the years, and how the majority of them have always ended up hurting the consumer at the end of it all. Being able to see where we as consumers are affected by these tariffs is very important to minimizing the impact that these tariffs can have on us.

What it does

Tarix is a web application that aims to give users more information on the products they are buying and how they are being tarifffed

How we built it

Tarix was built using Next.js and Flask for the web portion, I used a hugging face transformer to embed the vectors into my database locally with CUDA. I used a hosted instance of Postgres on Aiven for my database. I used the same the same embedding model hosted on hugging face, for embedding each prompt the user sends.

Challenges we ran into

I ran into quite a few challenges while working on this project. One Challenge I ran into was how I was going to do the embeddings for the retrieval augmented generation. I was unsure what to use as Models such as OpenAI can get expensive, I wanted to find a cheaper way of doing so. So I decided to embed locally which was really rewarding but somewhat challenging.