truify

Introduction/Summary

Truify is an agentic, open source project to help data and AI pratcitioners prepare training data. Truify provides CLI and web/GUI access to tooling enabling solutions to the following data preparation challenges:

• Bias (selection, recency, geographical, gender, channel/market)
• Compliance (GDPR, EU AI, SOC2, ISO 27001, HIPAA, CCPA)
• Privacy (PII, PHI)
• Missingness (incomplete, reverse-incremental)
• Imbalance (lack of adequate case vs. control in training data)
• Misalignment (cannot harmonize)
• Inaccessibility (unstructured/semistructured i.e. PDFs and photos)

Approach

Truify addresses each of these problems with known best-practice approaches as follows.

Problem	Approach
Bias	Data balancing through integrating weighting with public standards
Compliance	De-identification, deterministic filtering based on compliance policy, full auditability, including remediation and validation
Privacy/Missingness	Imputation w/”temperature” control (veracity/hallucination), full data synthesis
Misalignment	GenAI-enabled fuzzy matching based on observed records, automated harmonization

Benefits

Through these approaches, practitioners can achieve the following benefits:

• better accuracy
• better recommendations
• greater coveragegreater ROI
• improved sustainability
• lower risk
• greater efficiency
• greater trust
• improved transparency
• faster time to value

Getting Started

To see a sample streamlit application using the truify libraries:
OPENROUTER_API_KEY="your_api_key_here" python -m streamlit run code/main.py