AI systems grounded in database engineering

Santiago Andrés Aragón

AI Engineer Senior Database Architect

I build dependable AI systems on top of enterprise data. Five-plus years architecting and tuning Oracle PL/SQL and PostgreSQL databases now inform my work on deterministic RAG, semantic AST firewalls, and multi-agent orchestration.

Engineering in production

Proof over promises.

5+

Years in enterprise data

Oracle PL/SQL, PostgreSQL, and query optimization for mission-critical workloads.

1,085

Automated tests passing

Flagship Text-to-SQL RAG pipeline, exercised across security and dialect behavior.

245+

Adversarial cases tested

SQL injection and privilege escalation attack vectors validated against a deterministic AST firewall.

Maker→Checker

Multi-agent delivery

Structured orchestration with the Maker-Checker protocol and Orca ADE.

Selected work

Systems with a job to do.

More on GitHub
FLAGSHIP · AI / DATA2026

Text-to-SQL RAG

Natural-language questions become safe, executable SQL—with schema context and an AST firewall between the model and the database.

Dual-engine support for PostgreSQL 16 and Oracle 23c Free. Fail-fast handling reduced unreachable-database response time from 138.5 seconds to 4 seconds.

FastAPIpgvectorsqlglotOllamaOracle
View repository
TEST SUITE1,085 passing (89 skipped)
SECURITY CASES245 validated
FAIL-FAST4 seconds
Model output is untrusted input.
DEVELOPER TOOLING

PL-SQL Lens

A static analyzer and LLM-assisted utPLSQL test generator for understanding and testing legacy Oracle packages.

Oracle PL/SQLStatic analysisutPLSQL
AGENTIC WORKFLOW

Job Auto-Apply

An autonomous application pipeline that adapts candidate materials and cover letters to each role.

Multi-agentAutomationLLMs
FULL-STACK PRODUCT

Mr Games

A production e-commerce platform connecting a complete storefront experience to its operational backend.

CommerceBackendProduction

Try the safety boundary

A query earns its way through.

Choose an example to see how the demo pipeline treats a safe request and a hostile one. This is a client-side simulation; no query is sent to a server.

AST FIREWALL / SIMULATION LOCAL DEMO

inspect --example legitimate-query

INPUT“Which customers have overdue invoices?”
ALLOWED23 ms · simulated
SELECT customer_id, due_date
FROM invoices
WHERE status = 'overdue'
LIMIT 100;

AST valid · read-only · row limit enforced

Illustrative output only. The live application validates generated SQL before database execution.

Technical range

The model is one part of the system.

AI & LLM systems

RAG architectures · local inference · semantic AST guardrails · prompt design · multi-agent orchestration · Maker-Checker · Orca ADE

Data & databases

Oracle Database · PL/SQL · PostgreSQL · pgvector · SQL optimization · schema design · query safety

Backend & delivery

Python · FastAPI · Docker · Linux · CI/CD · automated testing · production debugging

Build carefully. Ship with evidence.

Let’s solve the hard part.