• 

  Screenshot of the ConUGPT

Inspiration

During enrollment periods, choosing the right class with the right teacher can be a hassle. Finding a professor that matches one's learning pace is easier said than done. The number of sources students usually have to consult makes the pressure of enrolling heavier. Now, what if all the needed information was one click away? We, the team Rush Hour 3, took matters into our own hands. Plus, from the GPT store which launched about 2 months ago, and thanks to a member on the team which has the ChatGPT+ account, we had the idea to create a GPT bot dedicated for Concordia students/general public.

What it does

It can provide general information on Concordia. But more specifically, it can provide information on the COMP and SOEN classes that new undergraduates can take including the class names, the teachers of the class and their respective ratings. Plus, we have a whole website for tutorials for those not familiar with LLM or question prompts.

How we built it

We used OpenAI API playground + GPT store to first configure with the knowledge base of the Chatbot, and then we used Replit for backend and Voiceflow for frontend to create the Chatbot linked with the API key, finally, we integrated it onto our main website(written in python ) as a script.

Challenges we ran into

The knowledge base is delicate, since we didn’t fully know how well the model can retrieve the information. So initially, the information regarding the classes were put into tables in pdf format. But, it turns out the model turns the table into a csv file internally, and inconsistencies appear. The model would give false answers, forget the previous knowledge or misunderstand the positions of the values inside the table.

Accomplishments that we're proud of

We figured out that the knowledge base is best integrated with txt files and the values separated by comma. This way, the model does not need to transform the values again for internal comprehension. This method is not perfect if the dataset was very large with complex writings/graphs, but for our project, the accuracy is sufficient enough. We managed to integrate it through replit and Voiceflow rather efficiently with the help of ChatGPT.

What we learned

Member 1:Shared databases on Github are very appreciated Member 2:Integrating different components & layouts of the website Member 3: Learned how programming works with OpenAI API and how it can lead to the next revolution of Programming & Building Projects

Usage of ChatGPT

We needed extensive help for the backend installation of the ChatBOT using Replit Including the basic framework for the code to work and how it is connected to the frontend (Voiceflow) using the API key.

Problems left

The model still could sometimes answer back to questions not related to Concordia, even though it is given clear instructions that it should only respond back if the relevant answers exist in the knowledge base. Part of this is due to lack of configuration from our end due to limit of time, but the OpenAI platform is also partially responsible for not recognizing and execute the instructions in the fullest extent.

Built With

• css
• github
• google-drive
• gpt
• html
• javascript
• openai
• playground
• python
• replit
• streamlit
• voiceflow