Know Your Vehicle

Inspiration

Modern cars are very complex. Fixing and maintaining them requires advanced knowledge, and owners manuals have become very large and complex. We want to simplify the ownership experience by allowing users to know their vehicles easily and be able to diagnose problems, work on them, and know when they actually need to spend the money to bring it to someone more knowledgeable like a mechanic.

What it does

Our project uses snowflake arctic LLMs to find relevant information in the owner's manual of a vehicle. From a user query, our app returns detailed information from the manual along with a summary of the retrieved information.

How we built it

We built our application using mainly a few notable functions. Then the manual content is transformed into vectors using: SNOWFLAKE.CORTEX.EMBED_TEXT_768('snowflake-arctic-embed-m', "content") as vector_embeddings, then This is saved in table "BRONCO_PAGE_CONTENT_WITH_VECT_EMB". Next when we get query from the streamlit app, Cosine Similarity is used to filter pages from the vector embeddings of pages and the top 3 pages are picked as context to pass through arctic model. The function VECTOR_COSINE_SIMILARITY(vector_embeddings,SNOWFLAKE.CORTEX.EMBED_TEXT_768('snowflake-arctic-embed-m','{query}')) is being used to do this. Once this is done, context is passed to Arctic through, SELECT SNOWFLAKE.CORTEX.COMPLETE('snowflake-arctic','{final_prompt}. Finally The response generated is being summarized using the function, SELECT SNOWFLAKE.CORTEX.SUMMARIZE('{data}').

Challenges we ran into

the first issue we ran into was exceeding a token limit. because of this, we had to lower the page count that our model was retrieving for a query.

Accomplishments that we're proud of

Finishing on time with a useful application! This app is genuinely useful to an average user. democratization of knowledge is very important to us and this application showcases the power of that. Car owners deserve to know how to work on their cars, diagnose issues, and save themselves time and money.

What we learned

We learned how easy it can be to leverage LLMs in snowflake. This really is a game changer. The amount of applications that can be built is truly amazing.

What's next for Know Your Vehicle

multiple vehicle support specific functionality sliders like vehicle mileage additional contextual data from forums

Built With

• coretex
• snowflake
• streamlit