_{Only those crazy enough to believe they can change the world are the ones who do.
I want to contribute to the world beyond my day-to-day work by leaving a trace of my nerdiness in public, so it can be useful to others.
May what I learn become a path others can walk to get further, faster, just as I once stood on the shoulders of giants.
I try to be a little better every day, treating setbacks as part of the process.}

Principal Data & AI Architect @ Mendel — building data and AI platforms.

Writing about agentic data systems on Medium — agents that debug their own pipelines, multi-agent architectures on AWS, modern data platform design.

Background in database systems through Universidad Tecnológica Nacional. Recognized at Hackathon 2025 - Harvard Health Systems Innovation Lab for building high-value health systems through AI.

• Agentic Airflow: Using AI Agents to Troubleshoot DAG Failures — Airflow failure context sent to an AI agent that reads code, proposes fixes, opens draft PRs, and notifies the team.
• Multi-Agents with Bedrock and Claude: One Supervisor and Three Specialists — a routed multi-agent architecture using AWS Bedrock and Claude models.
• Designing a Modern Data & AI Platform on AWS: From CDC to AI-Powered Analytics — an end-to-end AWS data platform design, from ingestion and transformation to observability, RAG, and AI-powered analytics.
• POC; Amazon Bedrock + S3 Vector — a practical Bedrock agent proof of concept grounded on documents through S3 Vectors.

I am contributing a cross-cloud set of Airflow integrations that turns a Dag into a control plane for managed AI agent runtimes. These operators bring the deployment lifecycle into the workflow itself: create an agent runtime, wait for it without occupying a worker, invoke or query the agent, publish updates, and clean up the infrastructure when the workflow is done.

• Amazon Bedrock AgentCore Runtime — create, wait for readiness, invoke, and delete an AgentCore Runtime.
• Vertex AI Agent Engine — create, retrieve, query, update, and delete an Agent Engine, with deferrable query jobs.
• Microsoft Foundry Hosted Agents — create and version, wait for activation, invoke through Responses or Invocations, and delete a Hosted agent.

The larger idea is bring your own agent: the agent keeps its framework, reasoning loop, models, and tool integrations, while the cloud service operates the managed runtime around it. Airflow does not replace either layer; it orchestrates the agent's operational lifecycle as part of a larger data or AI workflow. Each contribution includes hooks, operators, deferrable triggers, documentation, unit tests, and end-to-end validation against the real cloud service.

apache/airflow contributor Cross-cloud AI agent lifecycle integrations for AWS, Google Cloud, and Azure, plus data engineering operators

agentic-airflow-demo Practical examples for integrating AI agents with Airflow — AWS Bedrock AgentCore and GCP Gemini Agent Platform