How We Helped Precisely Enhance Their Data Integrity Suite with Production-Grade AI Agents

Manual, expertise-heavy workflows limiting AI readiness As enterprises accelerate AI adoption, the quality, consistency, and context of underlying data becomes mission-critical. Precisely's Data Integrity Suite, trusted by over 12,000 organizations globally, needed to evolve beyond existing capabilities to meet this demand.

Data teams were relying heavily on manual effort and specialized expertise to perform high-impact operations: writing quality rules from scratch, normalizing inconsistent data, enriching records, and cataloging assets. These workflows were slow, error-prone, and difficult to scale, creating a bottleneck between data integrity ambitions and AI-ready outcomes.

Precisely needed a way to automate these complex workflows intelligently, without sacrificing transparency, governance, or user control.

A config-driven AI agent platform on a proprietary foundational framework Our team architected and delivered four AI agents, Data Quality, Data Enrichment, Location Intelligence, and Data Catalog, deeply integrated into the Data Integrity Suite and its Gio™ AI Assistant. Rather than building isolated agents, we designed a robust foundational AI framework to power all of them consistently and at scale.

Our structured approach included:

• Foundational AI Framework: Built a declarative, config-driven framework on top of PydanticAI using a domain-specific language (DSL). Accepts per-agent configs for system prompts, guardrails, MCP server bindings, short-term memory, and LLM connections, instantiating stateless agents dynamically at runtime. New agents require zero code changes.
• AI Foundation Registry: Centralized all agent configurations in a single registry serving as the source of truth for every agent across the suite. The framework pulls from this registry at runtime to instantiate agents consistently.
• Powered by Claude on AWS Bedrock: The primary model is Anthropic's Claude Sonnet hosted on AWS Bedrock, providing secure, scalable, and compliant model access critical for Precisely's enterprise customer environments.
• MCP Server Architecture: Built dedicated Model Context Protocol (MCP) servers for each agent, exposing tools to read, create, and update assets within the Data Integrity Suite, enabling real action, not just responses.
• Vector Store on AWS S3 with Titan Embeddings V2: Built a continuous vectorization pipeline using Temporal that indexes all DIS assets into a vector store on Amazon S3. Powered by Amazon Titan Embeddings V2, agents can locate any asset through natural language queries with high accuracy.
• Knowledge Bases for Domain Intelligence: Developed knowledge bases seeded with Precisely's standard enrichment patterns and data quality rules accumulated over decades of domain expertise, giving agents a strong reasoning foundation.
• Bring Your Own LLM (BYOLLM): Designed the framework to be fully provider-agnostic. Customers can connect their own LLM via the UI, AWS Bedrock, Azure OpenAI, or GCP Vertex AI, and the framework routes execution through that connection seamlessly.
• Evaluation Framework on Datadog LLM Observability: Built a dedicated eval framework that maintains test scripts in Amazon S3, instantiates agents via the foundational service, runs queries at scale, and uses an LLM-as-a-Judge methodology to score outputs, all surfaced in Datadog dashboards for continuous quality monitoring.

Intelligent automation, governance, and a scalable AI foundation

• Reduced Manual Effort: Data teams can generate, review, and apply quality rules and enrichment actions through conversational interaction, eliminating the need for specialized technical expertise.
• Rapid Agent Expansion: The config-driven framework means new agents can be prototyped and deployed without engineering rework, dramatically reducing time-to-value for future AI capabilities.
• Enterprise-Grade Flexibility: BYOLLM support gives Precisely's customers full control over their LLM provider, meeting procurement, compliance, and cost requirements without vendor lock-in.
• Continuous Quality Assurance: The Datadog-integrated evaluation framework enables ongoing, automated assessment of agent behavior at scale, giving teams confidence in every deployment.
• Agentic-Ready Data at Scale: By combining intelligent automation with human-in-the-loop approvals, the agents help organizations build accurate, consistent, enriched data foundations ready for autonomous AI.

By combining a robust foundational AI framework with AWS-native infrastructure, including Claude Sonnet on AWS Bedrock, Amazon S3, and Titan Embeddings V2, and deep data integrity domain expertise, we helped Precisely transform complex, manual data workflows into intelligent, conversational, and auditable agent experiences. The result is a scalable foundation for the next generation of the Precisely Data Integrity Suite, enabling enterprises to achieve Agentic-Ready Data with confidence.