One of the main challenges we tackled was implementing video streaming for our rooms. At first, we thought about building a custom streaming solution from scratch, but it quickly became clear that time wasn’t on our side. We then looked into using LiveKit, which is a great tool for adding real-time audio and video streaming. However, setting it up was going to be a bit more complicated than we could manage within our project timeline.

That’s when we found Jitsi, which turned out to be a lifesaver! Jitsi offered an easy, ready-to-go video conferencing solution that fit perfectly with what we needed. It allowed us to set up a "Start Stream" feature for room creators and a "Join Stream" button for members without diving into too much setup. This way, we could deliver a reliable, user-friendly streaming experience without over-complicating things, staying on track while still meeting our goals.

Another challenge that we faced was figuring out the Gemini API. Traditionally we were skilled with building LLM based applications with Langchain, but to learn a new toolkit was challenging. Surprisingly the documentation and the video demos helped a lot more than we had expected. We did face trouble hitting the token quota when trying the "grounding" feature, but were able to build a good model regardless.

Having built a video streaming platform, one thing that was always in our mind was the issue of copyright and safety of customers of the platform on mind. We tried to brainstorm and build an AI based content monitoring system as a solution but that proved to be challenging, especially because the videos were being streamed. The limited time frame of the hackathon meant that we were unable to implement a solution for this but it remains our priority and a future improvement idea for our platform.