Inspiration
We created a chatbot to engage with and answer questions from prospective students interested in attending Drexel University. In addition to the general knowledge and language skills the chatbot possesses from the baseline OpenAI large language model training, we augmented its specific knowledge of Drexel University by scraping and structuring data from Drexel University's website. Then, we integrated these data with OpenAI's GPT-3.5 to create a responsive chatbot specially suited to its designed task. The frontend is a minimal, yet effective, chat interface that can be easily embedded into existing websites along with a larger scale and more completely interactive site provided by Azure's web deployment service.
What it does
Our chatbot leverages OpenAI's GPT-3.5 to answer queries about Drexel University in natural language. From admissions procedures to student life and current activities, it provides accurate, timely information based on data scraped from the university's website and updated regularly. These capabilities are easily integrated into any university webpage with just a single line of code and operate efficiently from a tokenized quantification billing scheme.
How we built it
The university-specific knowledge was drawn from the current information shown on Drexel University's website, which we scraped from the web and structured into digestible JSON packets using Python and Beautiful Soup. This information was indexed and integrated into the OpenAI GTP by search indexing to be available within seconds of each update. The frontend interface employs JavaScript and HTML to produce a simple, yet effective, chat interface that will be intuitively familiar to prospective students and their families..
Challenges we ran into
As novices, we found navigating Azure, and its multitude of configuration settings, to be difficult; this is where we spent 90% of our time. The value of CloudForce became immediately apparent and could greatly accelerate development of this and more complex projects.
Accomplishments that we're proud of
Getting anything successfully built using Azure's services for first-time users is something to be proud of. We are especially proud though that we attained our goal of creating a _ functional _ chatbot completed, which _ actually _ learned from the current, specially information we provided and updated into its training knowledgebase. Further, we feel that the combinations of these services is more useful (in some ways) than pure GPT4 or Google, which adds to our accomplishments.
What we learned
We learned a lot about Azure, not to mention the intricate details of the Drexel University admission process and current events in student life (via testing the chatbot)
What's next for Team 3
Feeling empowered by this technology and our successes this weekend, we are substantially more likely to consider applying large language models and OpenAI's capabilities, more generally, in future projects. It could be done, we did it, and we can do it again.
Log in or sign up for Devpost to join the conversation.