Selected projects

Systems you can explain and operate.

Work across open-source integrations, agent runtime orchestration and data platform design. The common thread is simple: make the architecture explicit, test the difficult path and leave behind something reusable.

Current work

From proof to working path.

Open source · Python

Agentic Airflow Demo

A working Apache Airflow proof of concept that sends failed DAG context to a containerized AI agent, produces a diagnosis, opens a draft GitHub pull request and posts a concise summary to Slack.

Why it matters

The same troubleshooting loop runs across AWS Bedrock AgentCore, Google Vertex AI Agent Engine and Microsoft Azure AI Foundry Agents.

View repository Read the walkthrough

Open source · Apache Airflow

Cross-cloud agent lifecycle orchestration

Native Airflow operators for creating, invoking, updating and deleting managed AI agent runtimes across three cloud providers.

Why it matters

Three provider contributions were merged upstream, making the lifecycle operations available from Airflow DAGs.

See contribution record

In progress · Apache Airflow

Oracle Cloud Infrastructure for Apache Airflow

Foundational OCI support for the Oracle provider: a native Airflow connection and reusable hooks for OCI Generative AI Hosted Applications and Hosted Deployments.

Why it matters

The open contribution adds four authentication modes, optional OCI SDK packaging, connection testing and native-client hooks that establish the base for follow-up lifecycle operators.

View open pull request

In progress · Product engineering

Morgan DataHub

A multi-tenant data platform proof of concept packaged as a reusable container, with tenant configuration and image versions managed as infrastructure as code.

Why it matters

The current design favors isolated tenant runtimes, reproducible deployments and explicit promotion of container image versions.

Architecture · Published

Modern Data & AI Platform

A reference architecture connecting operational data capture, transformation, governance and observability with an AI-ready serving layer.

Why it matters

The design treats reliability, metadata and security as part of the platform instead of adding them after the first workloads arrive.

Read the architecture