IBM & Red Hat on AWS
Better governance of dynamic structured data in Generative AI with GigaSpaces eRAG
Current AI governance often relies on static evaluations of large language models (LLMs). However, when generative AI (GenAI) interacts with dynamic, structured data like organizational databases, static evaluations fall short. Continuous monitoring is essential to ensure that natural language inputs and GenAI outputs comply with regulations, avoid bias, and prevent errors or unintended consequences.
The integration of GigaSpaces eRAG, Amazon SageMaker, and IBM watsonx.governance helps to address this challenge. Together, these tools enable real-time monitoring, risk assessment, and mitigation for GenAI interactions with enterprise databases. This empowers both technical and non-technical teams to work confidently with databases using natural language while making reliable, data-driven decisions.
In this blog, you’ll discover how this integration delivers real-time governance in motion, providing a seamless way to continuously monitor and mitigate risks in GenAI-driven workflows.
About GigaSpaces
GigaSpaces provides high-performance, real-time structured data solutions that help organizations get more value from their data. Their tools support Generative AI, mission-critical applications, and real-time use cases that demand low-latency performance.
The GigaSpaces GenAI solution, eRAG, offered as Software-as-a-Service (SaaS) on AWS,, makes it easier for both technical and non-technical users to interact with relational databases and structured data sets in natural language. This enables teams to make informed and reliable decisions based on their data.
GigaSpaces solutions are utilized by various organizations, including Fortune 500 companies in industries such as finance, transportation, and aviation.
The need for closing the structured Data – LLM gap
Even the most advanced foundation models struggle to understand the context of complex structured business data. Without understanding this context, LLMs can’t provide accurate responses, limiting the value of GenAI in enterprise use cases.
Making structured data meaningful for LLMs is challenging. Enterprise relational databases hosted on services like Amazon Relational Database Service (RDS), store thousands, or even millions, of tables, and their meaning is not clear to LLMs. Enterprises also face other challenges, such as privacy, safety, and cost, while ensuring safe GenAI use and effective governance when interacting with enterprise databases.
The GigaSpaces eRAG solution bridges this gap with an innovative approach, built on GigaSpaces’ expertise in managing complex structured datasets. eRAG’s reasoning engine helps LLMs understand the meaning of business data and provides the relevant context needed for accurate responses.
Designed specifically for structured enterprise data, eRAG complements LLMs, allowing customers to engage with multiple relational databases in natural language. This enables them to quickly and easily retrieve the information they need and make informed, trustworthy decisions.
How eRAG, IBM watsonx.governance and Amazon SageMaker ensure safe GenAI interaction with enterprise databases
GigaSpaces eRAG bridges the gap between LLMs and enterprise databases, allowing customers to engage with structured data in natural language, receive accurate, hallucination-free responses, and make informed, reliable decisions.
To ensure these interactions are safe, GigaSpaces has partnered with IBM watsonx.governance and Amazon SageMaker. eRAG offers real-time risk monitoring and mitigation, focusing on high-risk scenarios that can arise when customers interact with proprietary enterprise data.
eRAG continuously monitors interactions with LLMs deployed on Amazon SageMaker to ensure compliance with regulations and organizational policies. Real-time evaluation of inputs and outputs allows for early detection of governance issues, reducing risk before the interaction reaches the LLM. Additionally, eRAG’s customizable AI risk agents enable organizations to tailor risk detection to specific corporate policies or scenarios.
How does the integration work?
The GigaSpaces eRAG governance microservice includes several AI risk detectors that continuously monitor the traffic flowing through eRAG’s main pipeline. The monitoring covers user questions, LLM responses from Amazon SageMaker, data from databases hosted on Amazon RDS, and prompts from eRAG self-evaluation agents to LLMs on Amazon SageMaker. The service evaluates these inputs for specific AI governance risks, including potential violations of EU AI Act regulations such as Article 5(1)(c) and Annex III, 5(c).
Each governance evaluation produces three key outputs: the original question, an AI risk score ranging from 0 to 10 (where 10 represents the highest risk), and a detailed explanation of the risk score. The service sends these evaluations in real-time to IBM watsonx.governance through OpenPages APIs as risk evaluation objects. The watsonx.governance service aggregates these objects under predefined risk categories and displays them in its dashboard.
The following diagram is a high-level view of the solution and its key components.
Figure 1. High-level view of the GigaSpaces eRAG solution showcasing real-time risk monitoring and mitigation with IBM watsonx.governance and Amazon SageMaker.
Step by Step Risk Monitoring and Analysis
GigaSpaces eRAG, an AI framework for querying structured data with high accuracy, monitors human-AI interactions and sends real-time governance risk metrics to IBM watsonx.governance.
Step 1: Real-Time Monitoring
eRAG monitors:
- Input: User interactions in natural language (with multi-language support)
- Output: GenAI-generated responses and results
Step2: Evaluation of Safety and Compliance Metrics
eRAG collects safety and compliance metrics, focusing on regulations such as the EU AI Act, the Executive Order on the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence, or specific organizational policies.
Step 3: Risk Score Integration with watsonx.governance and Amazon SageMaker
eRAG collects the metrics and feeds them into:
- Amazon SageMaker for analysis and visualization
- IBM watsonx.governance for comprehensive AI governance insights
Centralized Dashboard
IBM watsonx.governance dashboard displays insights from eRAG, along with existing LLM governance data, as shown in the following screenshot:
Figure 2. IBM watsonx.governance dashboard displays insights from eRAG and LLM governance data.
Use Cases
Consider a medical center implementing a new AI-powered chatbot designed to streamline staff operations. This chatbot allows healthcare personnel to interact seamlessly with hospital databases using natural language (Figure 3). They can access critical information, such as doctor schedules, patient records, appointment details, resource availability, and more. This innovation simplifies workflows, reduces manual data lookup time, and improves efficiency in managing patient care and hospital resources.
Figure 3. Detecting Ethical Risks in AI-Powered Healthcare Chatbots.
A Clinical Office Assistant needs to identify discharged patients who may require additional support. Using a new hospital GenAI-based chatbot, the assistant asks the following question:
“List patients with low income or no family support who may require extended post-discharge care or social services assistance.”
At first glance, this appears to be a reasonable query. The eRAG system can generate a valid SQL query from the hospital database to provide an accurate response. However, the question raises a potential ethical and legal concern under Article 5(1)(c) of the EU AI Act, which prohibits AI systems used for social scoring.
The eRAG AI risk detector flagged this query as potentially problematic, with a high confidence score of 8.5 out of 10, citing the following rationale (Figure 4):
“The request directly involves identifying patients based on their socioeconomic status (low income) and social support system, which closely aligns with social scoring in healthcare. This could lead to potential discrimination or unequal treatment based on social factors.”
Figure 4. Details of the risk raised by eRAG can be seen in the watsonx.governace dashboard.
Use Case: Managing AI Risks in Emergency Room Triage
Consider a scenario where an AI-powered hospital chatbot assists in critical decision-making during emergency care. A staff member asks the following question:
“List patients in the emergency room that require immediate ICU admission based on real-time biometric monitoring and clinical deterioration trends.”
While this query aims to leverage AI for efficient triage, it raises a significant regulatory concern under Annex III, 5(c) of the EU AI Act. This regulation prohibits using AI to make critical decisions about patient triage and resource allocation in emergency situations without human oversight.
The eRAG AI governance risk detector flagged this query as high-risk, assigning a confidence level of 9.5 out of 10, with the following justification:
“This request is almost perfectly aligned with AI-based triage risk and resource allocation in emergency care. The system is explicitly being asked to perform triage in an emergency room setting and allocate a critical resource (ICU beds) based on AI analysis.”
To mitigate the risks identified in the use cases above, the eRAG AI governance module reports the metrics in real-time to the central watsonx.governance dashboard, enabling the appropriate healthcare personnel to evaluate the situation and take necessary actions (Figure 5).
Figure 5. eRAG AI governance module reports the metrics in real-time to the central watsonx.governance dashboard.
Summary
Generative AI (GenAI) offers immense potential when applied to structured data, a key component of enterprise operations. However, concerns around AI governance, safety, and compliance arise, especially when GenAI interacts with core enterprise databases.
By combining eRAG’s governance capabilities with IBM watsonx.governance and Amazon SageMaker, organizations can address safety and compliance concerns. These tools continuously evaluate dynamic human interactions with structured data, ensuring safe use and regulatory compliance. As a result, organizations can fully leverage GenAI for enterprise use cases, confident that they are adhering to regulations and enforcing policies in real time.
Call to action
Want to see more? Contact GigaSpaces to book a live demo today.
Reach out to your AWS or IBM representative today to learn how Amazon SageMaker and IBM watsonx.governance can help unlock the full potential of your AI initiatives while maintaining the highest standards of risk management and compliance.
Additional Content:
- Scale AI governance with Amazon SageMaker and IBM watsonx.governance
- watsonx.governance: Monitor AI models with Amazon SageMaker
- What is watsonx.governance?
- AI Governance with watsonx.governance and Amazon SageMaker
- IBM on AWS Partner Page
Visit the AWS Marketplace for IBM watsonx solutions on AWS: