NeuralAesthetics/ai-inbox-openenv

🧠 AI Inbox: Emotion-Aware OpenEnv Environment

Evaluating AI agents on emotional intelligence, prioritization, and decision-making under cognitive overload.

🚀 Overview

Modern email systems optimize for organization — not for human well-being.

Users today face:

• —📩 Information overload
• —😵 Decision fatigue
• —😡 Emotionally stressful communication

This project introduces an AI-powered inbox simulation environment designed to evaluate how well AI agents can:

• —Understand emotional tone
• —Prioritize important messages
• —Take appropriate actions
• —Reduce cognitive overload

💡 Key Innovation

This environment introduces an Overwhelm Score to simulate cognitive load, enabling evaluation of emotionally intelligent AI agents in high-stress communication scenarios.

🌍 Real-World Applications

• —Workplace productivity tools
• —Customer support automation
• —AI personal assistants
• —Mental load-aware systems

⚙️ Environment Design

This environment follows the OpenEnv specification.

🔁 Core API

• —reset() → Initializes inbox with predefined emails
• —step(action) → Executes agent action
• —state() → Returns current environment state

👀 Observation Space

The agent receives:

• —Email content (text)
• —Metadata (sender, timestamp)
• —Inbox summary
• —Overwhelm Score

⚡ Action Space

The agent can:

• —Classify emotion → angry / neutral / positive
• —Assign priority → high / medium / low
• —Decide action → reply / schedule / ignore

🎯 Reward System

✅ Rewards

• —Correct emotion detection → +0.3
• —Correct prioritization → +0.5
• —Correct action → +0.7

❌ Penalties

• —Ignoring urgent email → -1.0
• —Wrong emotional classification → -0.8
• —Poor decision → -0.5

🧩 Tasks

😌 Easy — Emotion & Urgency Classification

Identify tone and urgency. Scored using accuracy (0.0–1.0)

😐 Medium — Email Prioritization

Rank emails based on importance. Evaluated against ideal ranking.

😤 Hard — Action Recommendation

Decide best action for each email. Tests contextual and emotional reasoning.

🤖 Baseline Performance

Run the baseline agent:

python3 baseline.py

Example output: Baseline Score: 0.77

🐳 Deployment

This project is containerized and runs on Hugging Face Spaces.

To run locally:

docker build -t ai-inbox . docker run ai-inbox

📦 Installation

pip install -r requirements.txt

📁 Project Structure

ai-inbox-openenv/ │ ├── env.py ├── models.py ├── baseline.py ├── tasks/ ├── data/ ├── openenv.yaml ├── Dockerfile └── README.md

🧠 Why This Matters

Most AI environments focus on games or static tasks.

This project focuses on:

• —Human behavior
• —Emotional intelligence
• —Decision-making under stress

🏁 Conclusion

This environment provides a realistic benchmark to evaluate how AI agents:

• —Understand people
• —Manage priorities
• —Reduce cognitive load

✨ Future Improvements

• —Multi-user inbox simulation
• —Dynamic email generation
• —Advanced emotional modeling

🤝 Acknowledgements

Built for the Meta x Hugging Face OpenEnv Hackathon.