FAIR Data for Agentic AI
AI transformation, Components of Agentic AI, Amazon Sage maker Unified Studio, Data to Knowledge, FAIR Data Principles
Series: AWS Data and AI (15 articles)
- 2FAIR Data for Agentic AI This article
I am writing this article to provide insight of my latest speaking session in AWS Community Day Chennai 2026.
AI Transformation
AI has been transforming rapidly in recent years.
Few years back we were using traditional AI, then Generative AI and now it is Agentic AI.
Traditional AI
It uses existing data to predict and hep in decision making. For example: suppose we want to travel to country X. It provides details of the country and available airlines and predict temperature during our period of travel.
Generative AI
It is used to generate new images and texts based on the prompts we provide. For example: It can generate a customized travel itinerary for us in previous example.
Agentic AI
It can reason, plan and act to execute actions. In our example: it can generate customized itinerary and book the hotel accommodations and flight tickets as well based on our preferences.
What are AI Agents?
Autonomous or Semi-Autonomous software systems that can reason, plan and act to accomplish goals in digital or physical environments.
Components of Agentic AI
Each component of Agentic AI - LLM's, RAG, MCP and AI Agents interact with Data. Hence it necessary to ensure the FAIRNESS of Data in each stage.
LLM
During training of Large Data, Data sets must follow FAIR Data principles so that Data used for training is unbiased and data across available platforms are used.
RAG
During creation of RAG Data, Organization must ensure Data is unified, FAIR data principles adopted. This will help providing unified solutions across an organization. For example: we may have Jira for Support management, Confluence for artifacts, ServiceNow for incidents, etc. If the data from all these are not unified and not available with FAIR data principles adopted, then RAG will be biased to specific platform or channel of data. AI outcomes may not be responsible AI.
MCP
Model context protocol must be able to access the metadata across the platforms and tools using standard protocols. Data must be available based on authorization only.
AI Agents
AI agents must be able to access the data (metadata) without any technical issues due to protocols or data formats. Data must be available to AI agents based on Authorization only.
Amazon SageMaker Unified Studio
Amazon has introduced SageMaker Unified Studio, which supports Open Lakehouse. This will help us catalog all the data and create unified data storage solution. It can ensure AI applications are able to access unified data. Actual data is accessible in open table formats.
Data to Knowledge
A brief overview of how Data becomes an knowledge in an application.
- Data is ingested using Data ingestions tools.
- Data is processed for Data Quality Management.
- Parallelly Data modelling is done in which architects design database schema based on Business use case.
- Taxonomy or Ontology of data is created based Organization business use case.
- Data catalogued with meta data details.
Data to Knowledge - Pyramid
Below image shows how we can map our Data analytic pipeline to each stage in Data to Knowledge pyramid.
- During data ingestion it can be mapped to Data stage.
- During data processing it can be mapped to Information stage.
- During Data cataloging it can be mapped to Knowledge.
- Accessing Data by AI Agents can be mapped to Wisdom
F.A.I.R Data Principles
Findable
•Governance must be in-place to ensure meta data is controlled, meaningful and consistently persistent across the organization.
•Metadata and Data should be easy to find for both humans and computers.
Accessible
•Data must be accessible for authorized users only and metadata must be accessible for relevant Users and AI agents.
•The protocol to access must be open or easily identifiable for authentication and authorization procedure to access the data.
•Metadata must be available when if the data is no longer available.
Interoperable
•Metadata must follow standard and FAIR qualified formats so that it is identifiable across multiple systems and AI Agents
•Data must interoperate with applications or workflows for analysis, storage and processing.
Reusable
•Metadata must be reusable to build interconnected complex systems.
•Metadata and data should be well described so that they can be replicated or combined.
Is your DATA Agent Ready?
- Ensure the data follows the FAIR Data principles which ensures Data quality and Data Governance.
- In addition, ensure Data Lineage is maintained to identify the source of data for data quality scoring if required.
- All these principles help us build responsible AI solutions which are un-biased.
Series: AWS Data and AI (15 articles)
- 2FAIR Data for Agentic AI This article
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article