Winner — EchoForge Studio at Nebius Rebuild Hackathon
In a world where AI often feels like a black box, we set out to create something different — AI-driven voice agents with full transparency and user control.
What We Built
EchoForge Studio — a platform for building transparent, customizable AI voice agents. Unlike traditional voice AI systems that hide their decision-making behind opaque models, EchoForge gives users clear explanations of AI decisions and granular control over agent behavior.
Architecture
The system is built around three core principles — transparency, control, and quality:
-
Decision Transparency Layer — Every AI decision is logged with reasoning traces. When the voice agent routes a conversation, changes tone, or triggers an action, users can see exactly why and adjust the behavior.
-
Customization Engine — Granular controls for voice parameters (tone, pace, personality), conversation flow logic, escalation rules, and response templates. Users design agent behavior through an intuitive interface rather than writing prompts.
-
Real-Time Voice Pipeline — Low-latency voice processing using streaming ASR (automatic speech recognition), LLM-based intent understanding, and neural TTS (text-to-speech) for natural conversations.
Technical Stack
- Voice Processing — WebRTC for real-time audio streaming, Whisper for transcription
- Agent Logic — LangGraph for conversation state management and decision routing
- LLM Integration — Multi-model support with explainable output generation
- Frontend — React with real-time waveform visualization and decision trace UI
- Infrastructure — Nebius cloud for GPU-accelerated inference
Key Capabilities
- Explainable Decisions — Visual decision traces showing why the agent took each action
- Live Customization — Adjust agent behavior in real-time without redeploying
- Voice Personality Designer — Configure tone, speaking style, and emotional responses
- Conversation Flow Builder — Visual editor for designing multi-turn conversation logic
- Audit Trail — Complete log of every interaction with reasoning annotations
Why Transparency Matters
Most voice AI platforms treat their agents as black boxes — you get output, but no visibility into why. This creates trust issues, especially in enterprise contexts where compliance and auditability matter. EchoForge Studio proves that powerful AI and transparency aren’t mutually exclusive.
Acknowledgments
Thanks to the organizers at Sprint.dev and Nebius for hosting the Rebuild Hackathon and pushing for a more open, user-centric AI future.