Inspiration
We decided to gain some experience with AI by taking advantage of the Cloudforce-Microsoft AI track for this Hackathon. We also wanted to to provide our university, Salisbury, with a great resource, our chatbot, to incoming and current students regarding the many questions they may have.
What it does
GullGPT is a Azure based chatbot that answers Salisbury related questions for potential, incoming, and current students.
How we built it
We created a ingestion script which loads html, csv, and pdf files that we scraped and stored on our server into our Redis vector store (hosted on Azure) using Auzre OpenAI embeddings. We also built an update script to scrape new data so the chat bot can always have the most up to date information. We then used Lang chain and LangServe to to create to REST API services - one that summarizes chat history and another that responds to prompts using the Azure OpenAI GPT 3.5. Our client was built with HTML, Javascript and Tailwind. It gets summarized chat history from the summarization service to save context tokens and the requests the chat service to get a response.
Challenges we ran into
Setting up NGNIX was a big challenge as it was a new technology that none of us have used or anything similar to it.
Accomplishments that we're proud of
We are very proud of the overall product we managed to create during HOYA HACKS 2024, however we are especially proud of the UI for gullGPT, the chat summation feature which minimizes the cost and token use for each AI response, and the chat history where the user is able to reference previously revealed information, keeping the conversation personalized and relevant for that user.
What we learned
We learned how to use Langchain
What's next for Team#58 (gullGPT)
We will be integrating gullGPT onto www.salisburyanalytics.com, which is already a great student run site that provides students with great statistics about courses.
Log in or sign up for Devpost to join the conversation.