The problem Gymbros Connect solves

For people who are into gymming and healthy living, it is an absolute necessity to have a continuous support system
that keeps them motivated. It's even better if this support system is their friends. Our app solve this problem by providing a way for friends in different places and sectors of their lives to be a part of each others support system and push each other. It takes the competitve spirit of gymming and bring it to the web for enhanced connectivity. You can be in different parts of the world and still be competing with your BRO! Also for beginners and highly motivated people it is necessary to understand that only working out is not enough to be healthy. You have to mindful of what you eat, how you sleep, stress etc. Usually you have to get a coach for these kinds of information, but now you can leverage LLM's as well. This app provides a chatbot that runs on LaMini LLM and is fine-tuned with data from well known and authentic sources. It is like having your GYM-BRO by your side at all times. You can ask it anything about your health related queries.

Challenges we ran into

1. We decided to use auth0 for user authentication but the documentation for integrating with the latest version of
   Next.js was not up to date. Although athenticating users was easy, building protected routes seemed difficult. We had to go through the github repos of other programmers to solve ths problem.
2. Building a simple yet comprehensive UI was also intimidating. None of our team members have proper experience building frontend apps. We saw many similar apps and trid multiple layouts before finalising on one. This was the most difficult part indeed as we were on a timer and the interface is one the most important parts of the app. All the team mebers pulled together for this one.
3. Displaing live charts for data visualisation were difficult but thankfully we found chart.js after some research.
4. The most intimidating task was building the chatbot. Not only did we have to browse through multiple documents and and data to verify their authenticity and utility but we also had to extract them and and turn them into embeddings and then store them in a vector database. This was a risky task to undertake in a time constrained environment as we hand never done anything like this in such a short time. Also we had initially planned to use Llama-7b quantised model for the task, after the first day of the hackathon we realised it would be impossible for us to host such a thing for free. After one of the judges recommended we moved to a different model one of teamates undertook the task. Sleep deprived it beacame even more difficult. Finaly in the end we were able to fine tune a smaller version of Llama (LaMini) and host it on RunPod for free using a student ID.