Kairos

Inspiration

Education today is static. Whether it's a PDF or a video, the content doesn't care if the student is bored, confused, or excited. We wanted to build a tutor that feels 'human': one that looks at you, understands your struggle, and changes its teaching method on the fly. We were specifically inspired to tackle taboo subjects (like Sex Education) where students are afraid to ask humans questions, providing a safe, scientific, and adaptive space to learn.

What it does

KAIROS is a real-time adaptive learning platform powered by Google Gemini 3.

Personalized Curriculum: It generates unique learning paths based on user age and interests.

Emotional Intelligence: Using the webcam and Gemini Flash, it detects facial expressions (Confusion, Joy, Neutral) every few seconds.

Dynamic Adaptation: If KAIROS detects confusion, it stops, simplifies the explanation, and generates a visual aid or video. If it detects boredom, it gamifies the content.

Multimodal Generation: It doesn't just output text; it creates image prompts and curates videos relevant to the specific context of the conversation.

How we built it

We built a high-performance Real-Time Agentic Architecture:

Frontend: Angular 18 with Signals for reactive UI and direct webcam stream processing.

Backend: Python Django with Django Channels (WebSockets) for full-duplex communication.

The Brain (Gemini 3 Pro): Handles complex reasoning, lesson planning, and "Safe-Content" filtering for sensitive topics.

The Eyes (Gemini 1.5/2.0 Flash): Processes video frames efficiently to detect micro-expressions without high latency.

Orchestrator: A Python-based agent system that coordinates the Emotion Agent and the Content Agent to deliver a seamless experience.

Challenges we ran into

Our biggest challenge was Content Safety vs. Education. We wanted to teach sensitive topics (like Anatomy and Sexual Health) to teenagers, but standard AI filters would block these requests. We had to engineer advanced prompts and configure granular SafetySettings to allow for scientific, educational dialogue without triggering censorship blocks. Another challenge was Latency: synchronizing real-time video analysis with text generation required optimizing our WebSocket architecture to avoid 'race conditions'.

Accomplishments that we're proud of

We successfully implemented a Multi-Agent System where the AI 'changes its mind' based on user feedback. Seeing the interface react to a 'confused face' by automatically simplifying the topic and showing a video was a magical moment. We are also proud of creating a robust 'Demo Mode' that proves the concept works flawlessly even under network stress.

What we learned

We learned that Multimodality is the future of education. Text alone is not enough. We also mastered the nuances of the Google GenAI SDK, specifically how to balance cost/speed using 'Flash' models for vision and 'Pro' models for deep reasoning. We learned that AI needs empathy to be truly effective.

What's next for Kairos

We plan to integrate Gemini Live for voice-to-voice interaction, allowing students to interrupt the teacher just by speaking. We also want to expand our 'Taboo Topics' library to help democratization of information in developing countries.

Built With

• angular.js
• django
• gemini
• gemini-3-flash
• gemini-3-pro
• gemini-api
• python
• typescript
• websockets