Data First, GenAI Second: Building AI-Ready Data on AWS
Building NIS2 Intelligence for Spain: an AWS data platform combining governed data, deterministic intelligence and GenAI for evidence-grounded insights.
Series: Building a Data & AI Platform on AWS (1 article)
- 1Data First, GenAI Second: Building AI-Ready Data on AWS This article
Most GenAI projects start with the model.
I'm starting with the data.
I'm building NIS2 Intelligence for the Spanish market, using real regulatory and corporate data from Spanish public sources to build a trusted, traceable and AI-ready data foundation on AWS.
The principle is simple: build the data foundation first, then add intelligence.
From fragmented data to trusted data
Regulatory intelligence combines EU legislation, BOE/BORME publications, ENISA guidance, corporate data and technical evidence across multiple formats.
Before asking an LLM to reason about this information, I need to collect, normalize, catalog and preserve its provenance.
That's why I see this as a Data Engineering problem first.
Bronze → Silver → Gold
The foundation is a multi-zone data lake on Amazon S3.
| Layer | Purpose |
|---|---|
| Bronze | Preserve raw data and original evidence |
| Silver | Normalize, validate and enrich |
| Gold | Deliver trusted data for analytics and AI |
AWS Glue, AWS Glue Data Catalog and Amazon Athena provide the processing, metadata and analytics capabilities around this foundation.
The objective isn't another document repository.
It's an AI-ready data foundation.
Not everything needs an LLM
If structured data, SQL or business rules can answer something reliably, I don't want an LLM guessing it.
I use two complementary approaches:
- Deterministic intelligence
SQL · Business rules · Classification · Analytics - Generative AI
RAG · Interpretation · Explanation · Unstructured evidence
Calculate what can be calculated. Use GenAI for what requires interpretation.
This improves explainability, reduces unnecessary model calls and helps control cost.
Evidence before answers
Once the data foundation is reliable, Amazon Bedrock can power the Generative AI layer.
NIS2 Intelligence architecture: from governed data to evidence-grounded AI insights on AWS.
For regulatory intelligence, however, a convincing answer isn't enough.
Every important conclusion should remain connected to its evidence:
Claim → Evidence → Document → Source
I want to know not only what the system concludes, but why it reached that conclusion and what evidence supports it.
That's where metadata, provenance, retrieval and Generative AI come together.
The architecture
Data → Governed Platform → Intelligence → Evidence-Grounded Insights
The order matters.
I'm not building a data platform around an LLM. I'm adding Generative AI to a governed data platform.
What's next?
Next, I'll move from architecture to implementation with Amazon S3, AWS Glue, Amazon Athena, Amazon Bedrock, RAG and evaluation.
I'll share what I build, the trade-offs I encounter and the reasoning behind each architectural decision.
Build. Explain. Teach. Share.
Series: Building a Data & AI Platform on AWS (1 article)
- 1Data First, GenAI Second: Building AI-Ready Data on AWS 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