[VCCM] Very Cool Chat Machine

VCCM Landing Page

Inspiration

As students ourselves, we always face the problem of receiving a too diverse response while using chatGPT to assist us with school work.

This is why we wanted to create a custom trained chatGPT which answers our question in a more focused context.

What it does

VCCM: Very Cool Chat Machine will accept materials uploaded by the user and train the chatGPT AI model which is used to answer questions from the user thereafter. Authentication is done with our current favourite log in method: sgID!

Fun Fact

Although the login page seems to contain multiple other ways for users to log in, including via Google, Facebook, and GitHub, those buttons will disappear when users click on them so they will only be able to log in via sgID! :D

How we built it

We build our frontend using React framework. We have a running live server for sgID authentication at https://vccm-sgid.onrender.com/. And our frontend is deployed on netlify. For our custom trained chatGPT model we make use of mongoDB and python script that uses the langchain.chat_models library to import chatGPT as well as gpt_index library to customise the settings.

Challenges we ran into

We managed to complete our sgID authentication locally but faced many time-consuming issues when it comes to live deployment. In fact the end product is unable to show it's full potential due to the sgID issue that remains unresolved. Nonetheless, we still want to show the concept and our UI for our chat hence we decided to redirect users to our chat interface despite the error.

Accomplishments that we're proud of

We are proud that we managed to incorporate sgID to a high degree in our application through just reading the developer docs and especially with not much support from the community (specific errors not really google-able)

What we learned

We learnt how to make use of various different models that can read ZIP files and how to create various APIs that link to chatGPT to generate responses.