adqa 0.1.2

Autonomous Data Quality Agent: profiling, detection, explanation, and fix proposals for data quality issues.

ADQA: Autonomous Data Quality Agent

The intelligent, autonomous agent for high-performance data quality inspection, risk detection, and automated remediation.

Getting Started • Key Features • Documentation • Rust TUI • Contributing • Downloads

---

🧐 Why ADQA?

In the era of Data-centric AI, your models are only as good as your data. Yet, Data Scientists spend up to 80% of their time cleaning data.

ADQA solves this by providing an autonomous loop:

1. Observe: Deep multi-dimensional profiling of your dataset.
2. Orient: Detect complex risks like PII leakage, statistical bias, and structural anomalies.
3. Decide: Generate an execution plan with prioritized remediations.
4. Act: Heal the data autonomously or with human oversight.

🚀 Vision

ADQA combines a robust Python backend for seamless pipeline integration with a high-performance Rust-based TUI for interactive observability. It bridges the gap between fully automated data engineering and the critical need for human intuition in data quality.

📚 Documentation

Detailed guides, architecture deep-dives, and full API references are available at: Mohammad-Talaat7.github.io/autonomous-data-quality-agent

✨ Key Features

• 🔍 Multi-Source Ingress: Direct support for CSV, Parquet, Excel, SQL (Postgres, MySQL, etc.), S3, and 300+ SaaS sources via Airbyte.
• 🧠 Intelligent Profiling:
  • Structural: Automated type inference and null-ratio analysis.
  • Behavioral: Outlier detection (Z-score/IQR), skewness, and cardinality.
  • Semantic: ML classifiers identify PII (Emails, SSNs, CCs) and domain-specific types.
• 🚨 Hybrid Risk Detection:
  • Rule-based: Deterministic checks for drift, range violations, and duplicates.
  • ML-based: Advanced anomaly detection via Isolation Forests and bias identification.
• 🛠️ Autonomous Remediation:
  • Advisory Mode: Generate audit-ready reports of what should be fixed.
  • Automatic Mode: Fully autonomous healing (impute, drop, clip, mask).
  • Human-in-the-Loop: Interactive approval of fixes via CLI or TUI.
• 📜 Full Traceability: Industry-standard data lineage and execution traces for every transformation.

📦 Installation

Python Library & CLI

pip install adqa
# Or for full ML + Data Ingress capabilities:
pip install "adqa[all]"

Rust TUI

The TUI is distributed as a standalone binary. Install via cargo:

cargo install adqa-tui

Or download pre-compiled binaries from the Releases page.

🛠 Usage

Command Line Interface (CLI)

Quickly inspect any dataset:

adqa analyze my_data.parquet --mode advisory

Python API

Integrate into your training or ETL pipelines:

from adqa import ADQA, ADQAConfig

# High-performance profiling and detection
agent = ADQA.from_path("data.csv", config=ADQAConfig(execution_mode="automatic"))
result = agent.analyze()

# Access the healed dataframe immediately
clean_df = result.dataframe
print(result.summary())

🖥 Rust TUI

Monitor your agent's reasoning in real-time. The Rust TUI provides a zero-latency dashboard for exploring data lineages, trace events, and approving remediation plans.

adqa-tui

🤝 Contributing

We welcome contributions! Please see our Contributing Guide to get started with:

• Adding new Detectors.
• Improving the Scoring Engine.
• Enhancing the Rust TUI.

📄 License

ADQA is released under the MIT License. See LICENSE for details.