Data Steward Assistant - Agentic AI for Data Governance & Quality

Data Steward Assistant gives data stewards a conversational way to discover, profile, govern, and trace enterprise data. Powered by a LangGraph-based AI agent with a configurable large language model (Azure OpenAI, Google Gemini, or other LLMs selectable per session), the solution chooses the right tool for each request - catalog search, SQL execution, quality checks, scheduling, or lineage analysis - and returns results as tables, interactive charts, and downloadable PDF reports.

HCLTech's professional services team implements the solution in your environment, so you get a fully operational deployment tailored to your governance policies.

• Application deployment on a standard virtual machine (e.g., Amazon EC2) with a FastAPI backend, React web interface, and PostgreSQL configuration database
• Data source and catalog connections to Snowflake, PostgreSQL, Google BigQuery, Azure Synapse Analytics, Amazon S3, Azure Blob Storage, OpenMetadata, Zeenea, and other connectors as needed
• User and role configuration with persona-based workspaces (Data Steward, Data Engineer, Analyst, Guest) and role-based access to registered data assets
• Quality-rule generation and scheduling aligned to your governance policies, with documentation and knowledge transfer at handover

The AI agent searches your data catalog, runs SQL directly on connected databases (Snowflake, PostgreSQL, BigQuery), reads files from Amazon S3 and Azure Blob Storage, suggests PII classifications for steward review, performs data masking, and publishes approved rules and descriptions back to your metadata catalog. Every SQL statement the agent runs is captured in a SQL History view for full audit traceability.

The LLM proposes quality rules for selected tables across six governance dimensions - completeness, uniqueness, validity, consistency, timeliness, and accuracy - with CRITICAL to LOW priority levels. Rules are profiled in place on Snowflake or PostgreSQL to produce 0-100% pass rates without moving data, then published to OpenMetadata as official TestCases.

Excel and CSV files from Amazon S3 or Azure Blob Storage can be checked in memory for nulls, duplicates, date validity, string length, regex patterns, and allowed values.

A persistent scheduler supports interval or cron schedules with time-zone selection, storing jobs in the configuration database so they survive application restarts. Jobs can be run immediately, paused, resumed, or deleted, with a full run history of scores and health status. Optionally, results are saved to a PostgreSQL table, and for any rule below 100%, up to 100 sample non-conforming rows per rule are isolated with the failed rule name and detection time for targeted remediation.

The agent works with OpenMetadata, Zeenea, and other catalog tools through its MCP server. It lists tables and schemas, reads classifications and tags, suggests field descriptions (reusing existing descriptions where available), and writes approved descriptions back. Amazon S3 files can be registered in OpenMetadata as containers with an inferred column model.

Lineage from OpenMetadata, live-schema primary and foreign keys (including inferred joins for warehouses that declare none) is merged into one knowledge graph. Stewards can run impact and root-cause analysis, find bottleneck and unconnected assets, and measure how much of the data estate has sensitivity classifications. An LLM-derived business ontology can be reviewed, exported as OWL/RDF (Turtle, RDF/XML, JSON-LD, N-Triples), or published to OpenMetadata as a glossary.

AI-suggested classifications are clearly labelled for steward review. Ontology drafts stay unsaved until approved. Inferred relationships are never shown as declared facts. Glossary publishing defaults to a preview mode. Every action keeps the data steward in control, reducing manual effort while maintaining governance accountability.

• Find data issues earlier through automated quality checks across six dimensions with scheduled monitoring
• Reduce manual stewardship effort by letting the AI agent handle catalog searches, rule generation, and description writing
• Maintain full audit trails with logged SQL history and role-based access controls
• Leverage your existing infrastructure with connectors for Snowflake, PostgreSQL, BigQuery, Amazon S3, Azure Blob Storage, and metadata catalogs like OpenMetadata and Zeenea