A US National parks Information retriever chatbot

If you want to travel to a US national park, wouldn't you like a chat assistant who could find the information you are looking for by just entering a few words?

Comment

Inspiration

I have been meaning to visit the National Parks of the USA but haven't been to a lot. I see that people travel all around the year and visit different sites in the National Parks of the USA. These national parks are so important, beautiful, and valuable that I wanted to build an AI advisory system that could assist people in finding information regarding the specific park that they are going to.

What it does

The system gives you a textbox and you type in your query about a park that you are going to. For example, you type in that this is my second time visiting the Acadia National Park and I forgot where I camped. Could you tell me some of the camping sites there? The system will retrieve the answer from the context using RAG.

How we built it

I used the Snowflake, Streamlit, and llama3-70b model. I built it using two PDF files. The system first chunks the PDFs and stores them in the database. Then, using the ML model I categorize the name of the PDF files and give each of them a label so it is easy to find the answer based on the type of question being asked. The Steamlit app is the front end of the application where the user can select different types of models.

Challenges we ran into

Chunking the PDF files was challenging. It could take some time but now that the data from the PDF files is already in the database, it is bearable.

Accomplishments that we're proud of

An actual chatbot that fetches the correct answer about the park that you mentioned. Making this functional and getting a healthy output is the biggest accomplishment while doing this project.

What we learned

I learned that different machine learning models can give you different results and you have to perform an analysis of which model is giving a better output. If none are giving a good output then we have to fine-tune the model with our data. That hasn't been the case yet but if in the future there is a need, then we have to go to the fine-tuning route.

What's next for A US National Parks Information retriever chatbot

I have a couple of ideas for this application. First, I would like to build a mobile application that could do this. Secondly, an extension to Google Maps whenever the user is at or near a US National park and they have the Google Maps app open, they could have an option e.g. "Ask anything about this {NAME_OF_THE_US_NATIONAL_PARK}" and the user can type in the question and get an answer. Thirdly, this can be integrated with the National Park Service iOS mobile application.

Built With

Share this project:

Updates