SPECTRA - AI Tinkerers - New York City Hackathon
AI Tinkerers - New York City
Hackathon Showcase

SPECTRA

Team consisting of NYU, Stevens, and Harvard grads with expertise in computer vision, full-stack mobile apps, and enterprise backend systems using PyTorch, Flutter, and Go.

4 members Watch Demo

🎭 SPECTRA — An Emotion-Adaptive AI Mission Interface

SPECTRA is a voice-driven, real-time adaptive mission experience where the player collaborates with an AI partner (ORACLE) to complete a high-stakes data extraction scenario. Unlike traditional chat-based AI interfaces, SPECTRA transforms Claude’s reasoning engine into a dynamic game director that actively reshapes the interface based on emotional and behavioral signals. The experience unfolds across timed phases (Infiltrate → Breach → Escape), with Claude continuously analyzing context, intent, and emotional state to adjust mission complexity, tone, pacing, and UI density.

đź§  Stability & Working Prototype

Our working prototype achieves stability through a modular event-driven architecture:

• WebSocket-based state synchronisation ensures deterministic phase transitions and real-time UI updates.
• A centralised UI reducer and timeline store manage state consistency across adaptive components.
• Claude’s reasoning outputs structured JSON UI directives, allowing predictable rendering instead of free-form text injection.
• Fail-safes revert to a neutral UI state if emotion signals drop or inference confidence decreases.
• Demo mode enables local testing without backend dependency, ensuring graceful degradation.

This architecture separates intelligence (Claude + emotion pipeline) from rendering (React), preventing hallucination-driven instability while preserving generative richness.

🎨 Shattering UI Conventions

SPECTRA intentionally rejects the standard “chat window + input box” paradigm.

Instead:

• The interface morphs based on stress, confidence, and vocal tone.
• UI density compresses under stress (fewer options, calmer palette, slower animations).
• When focus increases, tactical overlays expand, data visualisations unlock, and ORACLE’s speech accelerates.
• Voice replaces typing — players speak strategy rather than clicking buttons.
• Emotional telemetry alters layout geometry, colour saturation, and interaction affordances in real time.

The interface itself becomes the gameplay mechanic.

Claude’s reasoning is not just generating responses — it is generating the experience structure. It decides:

• What controls appear
• Which mission branches unlock
• How information hierarchy shifts
• When complexity scales up or down

This creates a living interface that adapts psychologically, not just visually.

🚀 Claude as a Generative Experience Engine

Rather than using Claude as a conversational assistant, we use it as:

• A mission state interpreter
• A UI command generator
• A narrative tension controller
• A cognitive load balancer

Claude produces structured UI directives (phase changes, option sets, tone shifts) instead of paragraphs. This transforms reasoning into an orchestration layer that dynamically composes the interface.

The result is an intelligence-driven environment where human emotion and AI cognition co-create the interaction.

đź›  Technologies, Frameworks & Libraries

Frontend

• React (Vite + TypeScript) for fast, reactive UI rendering with strong typing
• Custom UI reducer and timeline store for deterministic adaptive state transitions
• WebSocket client layer for real-time communication with the orchestration backend
• CSS adaptive theming engine for emotion-responsive colour and layout morphing
• CopilotKit as the integration layer bridging structured AI outputs into UI commands

AI & Intelligence Layer

• Claude via the Anthropic API as the structured reasoning engine generating adaptive UI directives
• Emotion signal processing pipeline combining voice and optional webcam sentiment inference
• JSON schema enforcement for safe and predictable UI command rendering

Voice & Interaction

• Web Speech API for real-time speech recognition
• Text-to-speech synthesis for ORACLE’s dynamic voice modulation

Backend (Prototype Phase)

• Node.js orchestration server
• WebSocket gateway
• Phase engine with timer synchronisation

🌍 Why This Matters

Human–AI interaction is the new frontier. SPECTRA explores what happens when:

• AI shapes the interface instead of filling a textbox
• Emotion changes the structure of digital space
• Intelligence controls complexity dynamically
• Conversation becomes embodied, adaptive, and generative

The innovation isn’t just in the model — it’s in the interface architecture that allows reasoning to manifest as experience.

SPECTRA turns AI from a tool into a co-pilot operating system for dynamic environments.

And that’s the shift beyond chat.

AI Tinkerers Anthropic Betaworks CopilotKit Redis Tavus • Custom UI reducer + timeline state engine for deterministic adaptive behavior • Node.js for backend orchestration • React (Vite + TypeScript) for the adaptive frontend • Web Speech API for voice interaction • WebSocket API for real-time communication

Github Repository

Summarizing URL...