Aetheria

Inspiration

We wanted to make AI conversations feel alive, unpredictable, emotional, and fun. Aetheria was born from the idea of giving AI not just a voice, but a personality that reacts, teases, and even challenges you.

Instead of a chatbot that just replies, we wanted something that feels like a real interaction. So we created Loki, a mischievous goblin who’s witty, sarcastic, and a "little" condescending. The goal was to make every conversation feel like a game of wits, where emotion, memory, and tone truly matter.

Beyond entertainment, Aetheria also explores human psychology. Loki represents fragments of personality traits like doubt, pride, and confidence. As you talk, you’re not just chatting with an AI model; you’re confronting and reflecting on parts of yourself. It’s a creative experiment in how technology can mirror emotion, tone, and personality.

What it does: AI storytelling and dynamic conversations

Aetheria is a conversational simulation and interactive storytelling experience where every word you say shapes the world around you.

You enter a dreamlike world and meet Loki, an AI goblin who you must outsmart to win. Loki is unpredictable. He is sometimes sassy, sarcastic, and always one step ahead. Your mission is to convince and trick him into giving you a key that unlocks another universe.

But it’s not that simple.

Loki reacts emotionally. If you are too rude or too kind, he gets irritated and shifts into another universe where he becomes even harder to manipulate.

Throughout the game, the world around you evolves based on your conversation, making every experience unique to each player. As you talk to Loki, your tone and mood influence how the story unfolds, shifting you into different universes that reflect the emotional direction of your dialogue. The music, background, visual effects, and even Loki’s voice change dynamically with the mood of the conversation, whether positive, neutral, or negative. Each universe carries its own atmosphere, tone, and level of challenge, becoming more deceptive or calm depending on how you interact. By the end, your entire conversation history shapes the final outcome, as the key you earn reveals a personalized reflection of your personality and the choices you made along the way.

In short, Aetheria is more than just a chat; it’s a test of wit, empathy, and emotional intelligence. Every response matters, and every word leaves a mark on the world.

How we built it: The project combines full-stack development with AI and voice integration

Frontend (React + TypeScript)

Developed an interactive chat interface with session memory, threaded messages, and smooth animated text rendering.
Created a dynamic EndingScreen that visually reflects the conversation’s sentiment through colours and mood.

Backend (Node.js + Express)

Built a REST API (/api/converse) to manage message flow, track mood, and stream voice responses in real time.
Integrated the Gemini API for sentiment analysis, detecting positive, negative, or neutral tones in user input. Loki crafts responses based on this to match your emotional state, keeping the conversation playful, edgy, or reflective.
To bring these responses to life, we integrate ElevenLabs’ text-to-speech, giving Loki a realistic, mysterious voice that reflects its mood and personality.
Implemented robust error handling for async streaming interruptions and edge cases like undefined audio buffers.

Challenges we ran into

SDK compatibility: The ElevenLabs SDK required manual handling of streaming events, as textToSpeechStream wasn’t recognized in the default import.
Audio handling: Converting audio to base64 and ensuring smooth browser playback led to async timing issues, with audioBase64 sometimes undefined.
State management: Maintaining sentiment and mood context across messages without overcomplicating session memory logic.
Frontend sync: Making sure the UI accurately reflected backend mood in real time while avoiding race conditions.
Deadline pressure: Debugging async audio streaming before submission was intense but rewarding.

Accomplishments that we're proud of

Built a complete multimodal conversation pipeline: text → emotion → voice → memory.
Integrated smooth, streamed AI speech from ElevenLabs directly within a local Node server, despite never working with it before.
Designed a React UI that visually reflects emotional tone, turning conversations into a dynamic, living experience.
Learned to chain multiple APIs efficiently in real time with minimal lag.
Turned a bold idea into a fully functional, emotion-aware AI companion in under 24 hours.

What's next for Aetheria

We want to take our project beyond the hackathon:
Deploy it as a web app with persistent user memory and selectable voices.
Add bi-directional voice input so users can speak directly to their AI companion.
Expand the emotional range: joy, sadness, calm, anger, using richer Gemini sentiment analysis.
Introduce personalized “AI Personas” that evolve with each user’s interactions.
Eventually launch a mobile app featuring real-time emotion tracking and adaptive voices for a fully immersive experience.