Clarkie

Inspiration

We discussed solving problems like notifying students how crowded the cafe is, or the athletic centre (gym). Also, we thought of solving the problem of navigating the campus buildings for a new student. Thus, we considered building a generic solution for all these problems- an AI assistant- Clarkie. Clarkie is a smart AI assistant which answers all the questions related to the curriculum (course suggestion), building navigation, availability of space in the cafe, athletic centre etc.

What it does

Provide suggestions and solutions to the student queries.

How we built it

We web-scrapped Clark's public data and fed it to the Google vertex AI --conversation agents. After making the AI ready we integrated it into a web application. The application is now online https://clarkie.thinkandsoar.com/

Challenges we ran into

We couldn't train the dataset in our machine as none of us had GPU specs. So, for the prototype purpose, we had to choose some other service and it was a major challenge to accurately tune the AI. Also, lots of the Google documentation for the model was obsolete as they are transitioning from older versions to new versions. So, we really had to work hard make the bot up and running to this level.

Accomplishments that we're proud of

Everyone of us were new to LLM and this was the first project for us in this category. It is not perfect yet, but yeah we are proud as a team to try to mitigate a major problem for students in the university.

What we learned

Firstly, we enjoyed a lot to work as a team and thank you IEEE Student Branch at Clark for providing us with this opportunity. Our learnings are web scraping, data preparation to feed the model, on the way we also got to learn how LLMs work and last not but least teamwork.

What's next for Clarkie

Rather than using Google service, we plan to develop our model which will be heavily customized to Clark's data and could also provide real-time data using the camera surveillance at different locations in the university. To sum up, we plan to expand our dataset to images and video as well.

Built With

• canva
• django
• figma
• google-translate
• jupyternotebook
• postgresql
• python