Inspiration

Users who rent EVs for short term may not be familiar with their rented vehicle and hence will find it difficult if issues occur with their car

What it does

It provides a chatbot interface to allow EV drivers to ask questions to their EV manual

How we built it

We used unstructured to partition and chunk the pdf, Gemini-embedding-001 to generate embeddings, MongoDB to store the embeddings and maintain the vector search index for each manual, Google Cloud Storage to store screenshots of each page of the manuals, Streamlit as the frontend and Google Gemini-2.0-Flash as the LLM for the chatbot. The application was dockerized and deployed on Google Cloud Run

Challenges we ran into

Difficulty trying to find an efficient way to generate multi-modal embeddings. We solved it by generating text embedding and using the screenshots of the pages as reference (based on the user story)

Accomplishments that we're proud of

We have created a simple, easy-to-use interface that can help provide information and reassurance to EV drivers who may find it difficult to solve issues with their EVs

What we learned

We were able to learn and experiment with different embedding techniques, and how to use MongoDB to efficiently perform RAG activities

What's next for Electric Sherpa EV manual RAG

We would like to expand the number of EV models being supported (currently, only Hyundai Ioniq 5, Kia EV 6 and Tesla Model 3 are supported).

Built With

Share this project:

Updates