Track B — Vertical AI

Generative AI in the Enterprise — RAG & AI Agents

Build real enterprise AI systems — a grounded RAG search, then AI agents, then a multi-agent setup — all with no-code tools, on one deliberately messy simulated pharma company.

UC Santa Cruz ExtensionSeven live online classes + a final project. Hands-on labs and live, in-class homework evaluations throughout — your agent is graded against a held-out question set.
See classes open now
Taught by faculty atAlba — The American College of GreeceUC Santa Cruz Extension20+ years teaching · thousands of students trained
What this class is

A hands-on course that takes you from "what is generative AI" to a working multi-agent system in seven weeks — using no-code tools like Dify and Claude Cowork, not a programming language. It's not a theory seminar: every class you build on a single simulated company, Halcyon Therapeutics, whose data is realistically messy. You progress through foundations → enterprise search (RAG) → agents → production, and the architecture you learn outlasts the tool names that change every few months.

Who it's for — Knowledge workers who want to build serious enterprise AI systems — RAG, agents, and evaluations — without writing code. Ambitious and technical in spirit, but no programming required.

In class you work on a synthetic company corpus — a fictional business modeled on real-world challenges, so you practice on realistic, messy data, not toy examples.

What you'll be able to do
  • Build and tune a production-grade RAG system over messy enterprise data
  • Design chunking, hybrid search, reranking, and evaluation for retrieval quality
  • Build AI agents that write their own SQL over Supabase via MCP, with least-privilege design
  • Expose an agent as an MCP server and orchestrate a multi-agent system with delegation and guardrails
  • Evaluate agents against held-out question sets and surface data-quality issues instead of trusting surface numbers
  • Apply enterprise guardrails, governance, observability, and EU AI Act basics
The AI agents you'll build

You leave with A multi-agent system where a CEO orchestrator delegates to specialist sales and finance analysts over MCP.

These aren't demos — they're working agents you build hands-on and tailor to your own role, data, and custom spec.

Sales Analyst Agent (V0→V5)

An agent built iteratively that answers tabular, narrative, and compound sales questions over enterprise data — auditing for data-quality issues rather than trusting the surface numbers.

Financial Controller Agent

An agent that reconciles raw Halcyon sales data against the CFO's published quarterly results and surfaces the figures that don't match.

CEO Orchestrator Agent

A supervisor agent that answers from a scoped inbox knowledge base and delegates every sales and finance question to the specialist agents over MCP — a least-privilege multi-agent system.

The path

How you build them, class by class

1–2How LLMs actually work

Foundations

  • Neural nets, transformers, inference, tokens, context windows
  • Embeddings and cosine similarity; why models hallucinate
  • Prompting, chain-of-thought, and context engineering
  • Token economics and the enterprise AI landscape
3–4The spine of the course

Enterprise Search (RAG)

  • Build a grounded RAG system in Dify over messy enterprise data
  • Tune retrieval quality and apply advanced retrieval patterns
  • Synthesize facts across multiple source documents
  • Evaluation, guardrails, and the EU AI Act
5–6From retrieval to action

AI Agents

  • Agent architecture, tools, and the MCP protocol
  • Build a sales-analyst agent iteratively, then rebuild it in Claude Cowork
  • Give agents secure, least-privilege access to enterprise data
  • Agent patterns, failure modes, and self-improving agents
7 + FinalOrchestration, governance, ship it

Multi-Agent & Production

  • Multi-agent orchestration and the own-your-AI wave
  • Security, governance, observability, and the agentic enterprise landscape
  • Final: a multi-agent CEO assistant that delegates to specialist agents over MCP
  • Least-privilege design and safety-first data access
Your capstone

Your hands-on capstone: you play the CEO of Halcyon Therapeutics, facing rumors in email and Slack of a problem in the dermatology revenue line. You build a three-agent system — a CEO orchestrator whose knowledge base is only the inbox, which queries a Sales Analyst Agent and a Financial Controller Agent over MCP for the numbers it can't see. The data holds the problem; you have to make the agents surface the answer. You submit answers to 10 evaluation questions plus a short write-up — portfolio-grade work your certificate is based on.

Your certificate is based on what you build — not attendance.

Ready to build your AI agent?

A small, high-quality cohort for professionals — fully online, hands-on from minute one.

See classes open now