2026-AI-Development

Winter 2026 AI Development Training Series

This course was developed in partnership between the University of Chicago's Career Advancement Office and the University of Chicago's Data Science Institute.

Workshop: AI Development (4-part series)

This repository contains materials for a four-part workshop on AI development for advanced undergraduates (3rd/4th year).

Lecture 1: Foundations

Slides

• What LLMs are, how they work (tokens, context windows, temperature), and what they cost
• Hands-on with the OpenRouter API: sending your first programmatic LLM call
• Understanding the gap between "chatbot" and "AI-powered application"

Readings for Lecture 2:

• A Survey of AI Coding
• Effective Context Engineering for AI Agents — Anthropic's guide to designing context windows
• Context Engineering — Simon Willison's take
• Agent Best Practices — Cursor's guide to context engineering
• Welcome to Gas Town — LONG, only read as long as you are interested!
• Why ChatGPT Can't Be Trusted, Breaking: Marcus Weighs In, Let's Be Honest — Gary Marcus on AI limitations
• Claude Code Unpacked

Lecture 2: Building a System

Slides

• From idea to working prototype: vertical slices, MVPs, and the Crawl-Walk-Run framework
• Build a resume screening pipeline — one prompt that scores 130 resumes against a job description
• Understanding costs, latency, and model selection trade-offs in production

Readings for Lecture 3:

• Horizontal Slice vs Vertical Slice Programming
• AI Unit of Work — How to scope work for AI systems
• AI Blindspots — Where AI agents fail and how to design around it
• Scaling Agents — Cursor's insights on production-grade AI agents

Lecture 3: Making It Good

Slides

• Context engineering techniques that reliably improve LLM output: decomposition, grounding with citations, and few-shot examples
• Iterate on the Lecture 2 resume scorer until gold and silver candidates separate cleanly
• Learn why a well-engineered prompt can make even cheap models perform well

Readings for Lecture 4:

• Building Effective Agents — Anthropic's guide to agent patterns and best practices

Lecture 4: AI Agents & Tool Use

Slides

• Move from generating text to taking action: give LLMs tools that interact with the real world
• Build an agentic email outreach system — the agent scores resumes, decides outcomes, and drafts personalized emails
• Production considerations: failure modes, safety guardrails, ethical implications, and human-in-the-loop design

Running notebooks (per lecture)

Each lecture directory contains a Makefile, Dockerfile, and pyproject.toml.

From a lecture directory (e.g. lecture_2/):

• make build
• make notebook (starts Jupyter in Docker on port 8888)
• make interactive (drops you into a bash shell in the container)