Biosphere2 Ocean Data Chatbot

Inspiration

We were inspired by the opening ceremony, and the speech on AI. We wanted to create something that may be actually useful to researchers.

What it does

The Chat Bot takes in the most recent Ocean and Ocean SCADA data and feeds it to a Mistral AI. Users can interact with this through the web interface and ask the chat bot questions about the data.

How we built it

We are hosting Mistral 7B locally using Ollama. The web app interface is built using Streamlit. The web scraper is built in python using Selenium.

Challenges we ran into

This is all of our first hackathon, so we struggled with all of the new tools. AI in particular is a new subject for us! A large problem we still have is that a lot of the data is unprocessed, which gives the AI problems in analyzing the datasets. Our largest challenge was learning how LLM's functioned, and how to leverage that into something useful.

Accomplishments that we're proud of

We successfully created a chat bot that can respond to the user based off of the data sets using the web app interface. That seems like a very simple sentence, but we all worked on this for hours to accomplish this, and to see it finally working at 11:30pm was incredible.

What we learned

Going to the AI club workshop was a key moment for this team. They showed us how to use Ollama, and explained how AI works and how we could leverage it. We learned how to use Streamlit, improved our Python skills, and ate 3 boxes of Pizza.

What's next for Biosphere2 Ocean Data Chatbot

First, the chatbot only functions on one device, and cannot be easily moved to other devices due to the nature of the web scraper. Next, data processing is a key improvement needed for the chatbot. Clean data would allow the AI to have a much clearer understanding of the dataset. Improved hardware (and clean data) would allow us to scale the chatbot over the entire dataset.

Built With

• mistral
• ollama
• python
• selenium
• streamlit