Livvie

home page
live actions page
analytics page
review page

Inspiration

We watch a lot of livestreams where chat is constantly active, but most messages go completely unnoticed. Questions get buried, great moments are missed, and viewers slowly disengage.

That disconnect hurts both sides - viewers feel ignored, and streamers lose opportunities to build real engagement and grow their audience.

We built Livvie to solve this problem.

Livvie is a real-time audience intelligence tool that helps streamers understand what their audience is saying, respond at the right moments, and grow faster, all without breaking focus.

To take it a step further, Livvie suggests content ideas based on real-time trending tags and content niche to maximize engagement.

What it does

Real-time stream suggestions

Livvie continuously analyzes live chat and generates actionable suggestions, such as:

Content ideas to maximize viewership and engagement, influenced by chat suggestions and channel niche
Moderation actions like flagging or banning toxic users
Surfacing viewer questions that would otherwise get buried

Hands-off Assistant mode

Livvie can automatically act on suggested actions without human intervention. The streamer can enable automation for such actions in Moderation Settings, allowing them to focus on creating and engaging the stream.

Live audience analysis

Throughout the stream, Livvie tracks:

Engagement levels
Overall sentiment
Trending topics and tags emerging from chat

This gives streamers instant awareness of how their audience is feeling and reacting.

Persistence and Predictions for long-term growth

Livvie doesn’t stop when the stream ends. It stores session data and uses it to:

Suggest content ideas for future streams
Predict growth and identify target audience group
Help streamers understand what works over time
Support long-term brand and community growth

How we built it

Real-Time Streaming

Confluent Cloud Kafka: Handles chat messages, frame captures, and transcript events in real-time with SSL/SASL authentication
WebSocket: Pushes insights to the frontend instantly

Google Cloud Platform

Vertex AI (Gemini 2.5 Flash): Analyzes chat for confusion, monetization intent, and generates actionable recommendations at extremely low latency
Cloud Storage: Persists session data and captured stream frames
Speech-to-Text: Real-time transcription of stream audio
Translation API: Auto-translates chat messages
Natural Language API: Sentiment analysis on every message

Challenges we ran into

One of our biggest challenges was managing limited token usage over long development sessions, especially while testing real-time AI features at scale. One optimization was using a rolling window for messages to reduce the size API calls.

Accomplishments that we're proud of

We’re especially proud of turning a complex system into something streamers can actually use while live.

Dashboards and analytics are nice, but streamers need clear, immediate feedback - not more things to think about. Livvie focuses on surfacing the right action at the right moment.

What we learned

We learned that integrating new Google Cloud APIs and tools is surprisingly approachable. Reading documentation carefully and experimenting early made a huge difference.

What's next for Livvie

Next, we’re deepening Livvie’s integration with YouTube by connecting real-time chat intelligence with YouTube Analytics. This will tie audience sentiment, questions, and trending topics directly to metrics like retention, watch time, and subscriber growth - helping creators understand which moments actually drive performance and optimize future content.

Beyond YouTube, we plan to expand Livvie to additional livestreaming platforms, bringing the same real-time audience intelligence to more creators wherever they choose to stream.

There’s also major potential upside in leveraging the new ultra-fast Gemini 3 Flash model, paired with better prompting and model memory management, to push Livvie’s analytics deeper while making live insights truly real-time.