IShowStream

AI-Powered Twitch Chat Analytics

Comment

Inspiration

We kept noticing the same thing while watching streamers: chat moves so fast that important messages get buried. Questions go unanswered, donations get missed, and genuine fan interactions disappear in seconds.

We built IShowStream to help streamers actually understand their chat, not just moderate it.

What It Does

IShowStream analyzes Twitch chat in real time using 5 specialized AI agents:

  • SpamFilter blocks bots and spam
  • Priority ranks which messages matter most (1-10 scale)
  • Engagement predicts which messages deserve responses
  • Trend spots what's blowing up in chat every 5 minutes
  • Insight summarizes the vibe and key topics every minute

These agents work together — spam gets filtered first, then important messages get flagged, and trends/insights run in the background. Streamers see what actually matters instead of drowning in noise.

How We Built It

We used Python to coordinate the 5 AI agents (all powered by Gemini 2.0 Flash) and Go for fast backend services that handle chat ingestion and real-time updates. Everything runs on Google Cloud Run and stores data in Firestore.

The React dashboard connects via WebSocket so streamers see insights instantly as they happen.

Challenges We Ran Into

  • Getting 5 different agents to work together smoothly without slowing down
  • Processing thousands of messages per second in real time
  • Teaching the AI to understand messy chat (spam, emojis, inside jokes)
  • Keeping everything responsive during traffic spikes

Accomplishments That We're Proud Of

We built a 5-agent system that actually works in real time, deployed 4 different services that talk to each other seamlessly, and created a dashboard that makes complex AI analysis feel simple. The whole pipeline responds in under a second, even during heavy chat activity.

What We Learned

Building this taught us how to make multiple AI models work together as a team, not just individually. We learned that real-time AI is as much about coordination as it is about speed. Google Cloud Run turned out to be perfect for this kind of event-driven system.

Most importantly: understanding chat isn't about reading every message — it's about surfacing the signal through the noise.

Built With

Share this project:

Updates