LLM & Agent Builders
Mállíkana- og umboðssmiðir
Not just chat — agents that do things.
Book a free trialWhat they'll learn
- Apps on large language models Build real applications powered by LLMs.
- Grounding with RAG Use retrieval to ground AI answers in real documents instead of guesswork.
- Tools and agents Give an AI tools to use and build simple autonomous agents.
- Evaluate and guard Test an AI system and add the guardrails that keep it safe and reliable.
A working agent or RAG app that does a real task, with an evaluation of how well it works.
The 13-week journey
-
Sprint 1 — Beyond chat
LLM app architecture: prompts, context, structured output, function calling; building a first tool-using assistant. Ends with an app that calls a tool to get something done.
-
Sprint 2 — Grounded in data
Embeddings and retrieval (RAG): answering from real documents instead of guessing; chunking, search, and citing sources. Ends with a RAG app over a dataset they choose.
-
Sprint 3 — Agents & evaluation
Simple multi-step agents that plan and use tools; failure modes, guardrails and cost; building an eval set to measure whether it actually works. Safety and honesty throughout.
-
Week 13 — Showcase
Live agent demos with their eval numbers — and an honest account of where they still fail.
What we cover
Every topic, unit by unit — so you know exactly what your child builds and learns.
Beyond chat
- LLM app architecture: prompts, context, structured output
- Function calling / tool use
- Building a first tool-using assistant
- Handling and validating model output
Grounded in data (RAG)
- Embeddings and semantic search
- Retrieval-augmented generation
- Chunking documents and citing sources
- A RAG app over a chosen dataset
Agents & evaluation
- Simple multi-step agents that plan and act
- Failure modes, guardrails and cost control
- Building an eval set to measure quality
- Safety and honesty in agent design
Showcase & honest evals
- Presenting eval numbers, not vibes
- Live agent demos
- An honest account of remaining failures
A live agent completing a real multi-step task, with its tools, prompts and eval results on display.
The most current course in the school: building AI that does things, not just chats. Retrieval over real documents, tool use, and simple agents — with the evaluation and safety discipline that separates a demo from something real.
The hooks
Teen hook: “I built an agent that uses tools and answers from real documents — and I can prove how good it is.” Parent hook: “They learn exactly what the AI industry is hiring for right now, with safety and evaluation built in.”
Who it’s for
16–18s with solid Python (AI Engineer or AI App Builders + placement). Thrives: builders and the genuinely curious about how modern AI products are made.
Outcomes — by the end, students can
Build an LLM app with structured output and tool use; implement retrieval over documents and cite sources; build a simple multi-step agent; design guardrails and control cost; create an eval set and report results honestly.
Tools & compliance
Python, school-managed LLM API access with caps, vector-search tooling, own accounts at 16+, public/open datasets; safety, disclosure and evaluation norms enforced.
Where this course fits
Builds on AI Engineer; a strong feeder into the Launchpad capstone.
Parent questions
Is agent-building real and current?
Yes — tool-using agents and RAG are exactly what the industry is building right now; students learn the real patterns and how to evaluate them.
What do they need first?
Comfortable Python — AI Engineer or AI App Builders plus placement. We build the AI-specific parts here.
Do they just rely on AI to build it?
They use AI assistance like professionals — with review and evaluation — and own every line that ships.
The first lesson is a free trial.
Book a no-commitment trial — pay nothing if it's not a fit.
Book a free trial