Director of Cloud Data Engineering shares how he’s using CustomGPTs to enable engineers

Subash D'Souza is the Director of Cloud Data Engineering at California State University. He's also the founder of Data Con LA and Data 4 Good. And he has held leadership positions at Headspace and Warner Bros.

We sat down with Subash to learn how he's enabling engineers while maintaining governance and reviews. Here's what he said.

A career in data

I'm the founder of Data Con LA and Data 4 Good, a public-private nonprofit think tank that works with organizations to tackle social problems using data, including issues such as homelessness and digital equity, while advancing technology innovation, data and analytics, and public-private partnerships.

I'm also currently the Director for Cloud Data Engineering at California State University, Office of the Chancellor. I serve on the Linux Foundation as a Data Governance PMC member. And before this, I was Head of Data Engineering at Headspace and Director for Big Data & Ops at Warner Bros.

Leading data engineering and AI

I lead the Data Engineering and AI/ML organization within the California State University Chancellor’s Office, supporting data initiatives across the CSU system.

We explore how AI could improve the student journey from application through graduation. By building on our existing data platform and student data, we aim to use AI to identify patterns that improve application conversion, recognize students who need earlier support, and ultimately improve retention and graduation rates. This shifts us from simply reporting what happened to proactively using data and AI to help our campuses make better, earlier decisions that support student success.

Our cloud-based architecture uses AWS, Databricks, Snowflake, and analytics platforms to build scalable data pipelines and products. My team supports a complex, multi-campus environment where we develop shared solutions, establish engineering standards, and enable campuses to build and deploy their own data products through our Data Center of Excellence model.

How AI can enable engineers while maintaining review and governance

I introduced AI directly into our engineering enablement model through the CSU Data COE CustomGPT.

Previously, developers relied primarily on documentation and engineering leads for guidance and troubleshooting. We created a CSU-specific AI mentor grounded in our data engineering practices and documentation to help developers understand our architecture, development standards, and workflows.

It helps in three ways:

• For knowledge, the CustomGPT gives engineers immediate access to CSU-specific guidance on our architecture, standards, and development processes instead of searching through documentation or waiting for an expert.
• For development, it helps engineers understand established patterns and navigate our Databricks, Git, testing, and deployment workflows.
• For troubleshooting, engineers can describe an error or issue and use AI to identify likely causes and potential next steps.

This gave teams faster access to consistent technical guidance, reduced dependency on individual experts, and helped us scale data engineering knowledge across CSU campuses while keeping our established review and governance processes in place.

Why AI accelerates while humans remain accountable

So, we increasingly use AI to support activities such as code development, troubleshooting, documentation, testing guidance, and helping engineers navigate our CSU data engineering standards through tools like the CustomGPT that I mentioned above.

However, AI has not yet replaced human judgment in complex architecture, security, or production decisions. We've also found that AI struggles with questions depending on highly specific context or rapidly changing technical documentation.

For example, it may suggest a technically valid approach to a deployment or security issue without fully understanding our specific AWS, Databricks, Snowflake, access-control, or approval requirements.

That's why architecture decisions, production deployments, security and access decisions, and final code reviews remain explicitly human. Our development process requires peer review, testing, end-to-end validation, and defined approvals before code progresses toward production.

We use AI to accelerate engineering and improve consistency, while humans remain accountable for decisions that affect security, architecture, governance, and production systems.

Why CTOs must build governance before integrating AI

Every CTO should understand that AI transformation is not primarily a technology problem — it is a people, knowledge, and governance problem.

AI can dramatically accelerate how teams learn, develop, and solve problems, but its value depends on trusted organizational knowledge and clear human accountability.

Honestly, I wish I had recognized earlier that the value of AI depends heavily on the quality and organization of the knowledge behind it. For example, with our CustomGPT, we could have avoided some early experimentation by focusing sooner on curating the right knowledge, defining where AI should and should not be trusted, and establishing clear human checkpoints.

My biggest lesson from this was that you shouldn't just roll out an AI tool. First, you need to build the knowledge, governance, and human accountability around it.

Why CTOs must invest in both AI and their people

I advise CTOs to treat AI as an accelerator, not a replacement strategy.

Start with practical use cases where AI removes friction, improves developer productivity, and makes organizational knowledge more accessible. But keep humans accountable for architecture, security, and production decisions.

Equally important, invest in your people. Our experience with the CSU Data COE shows that combining AI with hands-on learning and cross-team collaboration builds stronger technical capability.

The organizations that succeed with AI will pair rapid experimentation with strong engineering discipline, governance, and measurable outcomes.

Follow along

You can follow Subash D'Souza's work on LinkedIn. And check out Data Con LA and Data 4 Good.

More expert interviews to come on The CTO Club!