AI Engineer for secure, evidence-driven systems
I build and ship AI agents and applications that are observable, policy-controlled, and provably correct — not just prompted and hoped for. Before this: 11+ years operating safety-critical systems in regulated nuclear medicine. Security isn't a specialty I bolt on afterward; it's how I was trained to build things that can't fail quietly.
$ whoami
> César Darío Aguilar
$ role
> AI Engineer
$ stack
> RAG · LangChain · Bedrock · Policy Engines
$ status
> available for remote AI Engineering roles
AI Engineering
LLM agents, RAG, evaluation, and production deployment. FastAPI, LangChain/LangGraph, Bedrock, pgvector.
Security & Verification
The control layer that decides whether an agent's work is actually correct: policy engines, deterministic quality gates, threat modeling.
Cloud Infrastructure
AWS, GCP, Terraform, Docker, CI/CD — the operational backbone every AI system needs in production.
Selected work
Featured projects
agent-orchestrator-soc
Multi-agent SOC triage system (LangGraph supervisor-worker) that turns a raw alert into a severity-scored triage report — real threat-intel enrichment (VirusTotal/AbuseIPDB/OTX), a human-approval gate before any High/Critical report closes, and a regression eval (5/5) to catch prompt/model regressions. Deployed live and triggered via n8n, validated end-to-end against real Wazuh/Elasticsearch alerts.
LangGraph · FastAPI · Ollama · ChromaDB · n8n · MITRE ATT&CK
aigis-control-plane
Evidence-driven control plane for AI coding agents — deterministic policy engine, Docker-isolated execution, quality gates the agent can't write to, and 8 scripted security evals. Found and closed its own P0 scope bypass; post-fix benchmark 24/24 PASS against Claude in Docker, $0.040 cost-to-pass.
Python · Pydantic · Docker · Policy Engine · Claude API
local-rag-second-brain
100% local RAG over an Obsidian vault (markdown-aware chunking, Ollama, ChromaDB) so hundreds of technical notes and lab writeups become queryable and cited — nothing leaves the machine. The foundation project the other two reuse.
Ollama · ChromaDB · LangChain · Python