Abstract
A self-healing data system applies generative governance to automate data validation at an ingestion point. The system continuously observes raw incoming data streams to infer underlying business rules, data relationships, and structural schemas. Based on these observations, an artificial intelligence (AI) model dynamically generates a machine-readable data contract. When an incoming record violates the data contract, the system routes the anomalous data and the corresponding data contract to a large language model (LLM) for data cleansing. The LLM autonomously resolves formatting inconsistencies, fixes typographical errors in structural keys, and normalizes values to reconstruct a compliant data record. Following this generative correction, the system passes the reconstructed data record to a deterministic programmatic validator. The programmatic validator evaluates the reconstructed data record against the data contract before final database ingestion. This automated validation sequence processes irregular input formats, reduces manual engineering overhead, and minimizes separate database cleanup efforts.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
N/A, "Generative Data Contract Discovery and Large Language Model Self-Healing for Ingestion Validation", Technical Disclosure Commons, (October 20, 2026)
https://www.tdcommons.org/dpubs_series/12336