Welcome & Overview

Welcome to ESDS Advanced, an applied training program from the North Carolina Institute for Climate Studies (NCICS). Where the foundational ESDS course builds cloud-native scientific programming skills, this course turns to the cutting edge of scientific practice: agentic cowork tools, agentic systems, and deep learning.

Why this course

The cutting edge of science now depends on skills in three areas — agentic cowork tools, deep learning, and agentic systems. Together they unlock:

Faster development of scientific projects.
Broader, more rapid synthesis of large amounts of information.
Higher-performing models, and new ways to get mechanistic insight from them.
Better tools for communicating with non-researchers.
New ways to connect science with impact in other domains.

Goals

By teaching these skills in an applied setting, the course aims to help you:

Do faster scientific development, and take on larger-scoped projects in the same amount of time.
Practice good scientific agency — understanding the long-term risks of delegating to agents, and managing personal and scientific risk.
Maintain high scientific standards and quality while using these tools.
Build non-technical tools that let stakeholders access your science.
Build higher-performing models by leveraging deep learning in your domain, and use those models to inform mechanism.

What you'll be able to do

Specifically, by the end of the course you should be able to:

Generate a project report in collaboration with Claude Cowork.
Tell the difference between quality and AI slop — and engineer for quality.
Use coding agents for scientific programming while validating the results.
Manage and orchestrate multiple parallel workstreams effectively.
Use and build Claude Code skills.
Build a REACT loop and a full agentic workflow, connect it to data, and validate, monitor, and launch it for yourself and others.
Understand the basic classes of deep learning architectures and what they are good for.
Build and train your own deep learning model on scientific data, and pull, use, and fine-tune open-source models.
Evaluate the performance of agentic systems and deep learning models.
Stay informed on new tooling, architectures, patterns, and ideas.

How the course is organized

The course runs over three weeks, each a self-contained module that also feeds a shared team project:

Week 1 — Agentic Cowork Tools. Using Claude and Claude Cowork, then scaling up to Claude Code. Open to everyone.
Week 2 — Agentic Systems. Building REACT loops, harnesses, skills, and full workflow systems.
Week 3 — Deep Learning. Building, training, and evaluating models on scientific data.

Use the sidebar to move through the material at your own pace.

Welcome & Overview

Why this course

The cutting edge of science now depends on skills in three areas — agentic cowork tools, deep learning, and agentic systems. Together they unlock:

Faster development of scientific projects.

Broader, more rapid synthesis of large amounts of information.

Higher-performing models, and new ways to get mechanistic insight from them.

Better tools for communicating with non-researchers.

New ways to connect science with impact in other domains.

Goals

By teaching these skills in an applied setting, the course aims to help you:

Do faster scientific development, and take on larger-scoped projects in the same amount of time.

Practice good scientific agency — understanding the long-term risks of delegating to agents, and managing personal and scientific risk.

Maintain high scientific standards and quality while using these tools.

Build non-technical tools that let stakeholders access your science.

Build higher-performing models by leveraging deep learning in your domain, and use those models to inform mechanism.

What you'll be able to do

Specifically, by the end of the course you should be able to:

Generate a project report in collaboration with Claude Cowork.

Tell the difference between quality and AI slop — and engineer for quality.

Use coding agents for scientific programming while validating the results.

Manage and orchestrate multiple parallel workstreams effectively.

Use and build Claude Code skills.

Build a REACT loop and a full agentic workflow, connect it to data, and validate, monitor, and launch it for yourself and others.

Understand the basic classes of deep learning architectures and what they are good for.

Build and train your own deep learning model on scientific data, and pull, use, and fine-tune open-source models.

Evaluate the performance of agentic systems and deep learning models.

Stay informed on new tooling, architectures, patterns, and ideas.

How the course is organized

The course runs over three weeks, each a self-contained module that also feeds a shared team project:

Week 1 — Agentic Cowork Tools. Using Claude and Claude Cowork, then scaling up to Claude Code. Open to everyone.

Week 2 — Agentic Systems. Building REACT loops, harnesses, skills, and full workflow systems.

Week 3 — Deep Learning. Building, training, and evaluating models on scientific data.

Use the sidebar to move through the material at your own pace.