North Carolina Institute for Climate Studies

ESDS Advanced

Applied skills in agentic cowork tools, agentic systems, and deep learning — so you can do faster scientific development, build higher-performing models, and connect Earth systems science with impact in other domains.

4 modules
15 lessons
Start the course
Part 1

Getting Started

What this course is, how it runs, who it is for, and the team project that ties the three weeks together.

  1. Welcome & Overview

    Why agentic cowork tools, agentic systems, and deep learning matter for NOAA/NCICS science — and what you will be able to do by the end.

  2. Logistics & Format

    Three weeks of mornings for lecture and practice, afternoons for project work, with weekly and final presentations.

  3. Prerequisites

    What you need before each week — from "no prerequisites" for the first two days to scientific programming experience for deep learning.

  4. Project Component

    Team projects that build real NOAA/NCICS tools, drawing one piece from each of the three modules into a final product.

Part 2

Week 1 — Agentic Cowork Tools

Open to everyone: how LLMs and agents work, then using Claude and Claude Cowork for research, reporting, and scaling up to Claude Code.

  1. Overview & Objectives

    The week at a glance: from the fundamentals of LLMs and agents to running parallel Claude Code instances with remote MCP.

  2. How LLMs & Agents Work

    Days 1–2: a working mental model of large language models, context, and what makes a system "agentic."

  3. Working with Claude & Claude Cowork

    Days 1–2: research, reporting, brainstorming, red-teaming, and admin — plus skills and context-engineering strategies.

  4. Claude Code at Scale

    Days 3–5: from the basics of Claude Code to parallel instances with remote MCP, including cross-language codebase conversion.

Part 3

Week 2 — Agentic Systems

Building agentic systems from scratch: REACT loops and harnesses, good skills, and full workflow systems with humans in the loop.

  1. Overview & Objectives

    Prerequisite: Week 1. What you will build this week, from a first REACT loop to a monitored, guard-railed workflow.

  2. REACT Loops & Harnesses

    Days 1–2: an introduction to agentic systems — building a REACT loop and a simple harness around it.

  3. Building Good Skills

    Day 3: what separates a reliable, reusable skill from a brittle one, and how to build them.

  4. Agentic Workflow Systems

    Days 4–5: human-in-the-loop, monitoring and observability, guardrails, cost, autonomy, and memory.

Part 4

Week 3 — Deep Learning

Building, training, and evaluating deep learning models on scientific data, then leveraging and improving open-source models.

  1. Overview & Objectives

    Prerequisite: Week 1 plus scientific programming experience. From DL fundamentals to validating open-source models.

  2. DL Architectures & Training

    Days 1–2: common deep learning architectures and what they are good for — then build and train a PyTorch model on AWS GPUs.

  3. Open-Source Models & Evaluation

    Days 3–5: pull, use, and fine-tune open-source models and architectures, and validate and evaluate the results.