ARIA: Enabling Self-Improving Agents to Learn at Test Time With Human-In-The-Loop Guidance

ARIA is an LLM agent framework designed specifically to continuously learn updated domain knowledge at test time through human-in-the-loop guidance.

Overview

ARIA assesses its own uncertainty through structured self-dialogue, proactively identifying knowledge gaps and requesting targeted explanations or corrections from human experts. It then systematically updates an internal, timestamped knowledge repository with provided human guidance, detecting and resolving conflicting or outdated knowledge through comparisons and clarification queries.

Core Components

• Knowledge Repository: Manages the storage, retrieval, and updating of domain knowledge
• Intelligent Guidance Solicitation (IGS): Evaluates agent confidence and formulates targeted queries
• Human-Guided Knowledge Adaptation (HGKA): Incorporates human feedback into the knowledge base

Installation

# Clone the repository
git clone https://github.com/your-username/aria.git
cd aria

# Install dependencies
pip install -r requirements.txt

# Install in development mode
pip install -e .

Usage

Basic usage example:

from aria.src.core.aria_agent import ARIAAgent
from aria.src.config.config import ARIAConfig

# Initialize configuration
config = ARIAConfig()

# Create ARIA agent
agent = ARIAAgent(config)

# Process an instance with human-in-the-loop
prediction, reasoning = agent.process_instance(instance_data)

For a complete example of using ARIA for Customer Due Diligence (CDD) tasks, see the examples/cdd directory.

## License

[MIT License](LICENSE)