Imagine you've just joined Brightpath, a fast-growing online-learning startup. Forty people, thousands of students, and a problem that's getting worse every week: students fire off questions faster than the team can answer, the course material is scattered across PDFs and wikis, support tickets pile up, and every lesson illustration is made by hand. The founder pulls you aside: "Build us an AI assistant that can actually help. Start small, but make it real."
That assistant is Sage, and building it — properly, layer by layer — is what you'll do for the next seventeen sections.
Why one product beats twenty demos
You could learn each topic with a throwaway example — a toy chatbot here, an unrelated RAG script there. The problem is that nothing connects, so every lesson starts from zero and the skills don't stack. By instead growing one product, each new capability plugs into everything you've already built. The prompt you write in Section 5 becomes the brain of the API app in Section 6, which gains memory of Brightpath's docs in Section 8, learns to use tools in Section 10, and finally ships to real users in Section 14.
Sage starts life as a single API call that can barely hold a conversation. It ends as a multimodal, agentic, retrieval-grounded, evaluated, deployed system that answers from Brightpath's knowledge base, uses tools, generates content, and runs in production. Same product — seventeen layers thicker.
Sage's four faces
Sage isn't one narrow bot — it grows four distinct capabilities, each unlocked by a hands-on project. Conveniently, these four faces map onto the kinds of GenAI products you'll meet everywhere in the wild, so the skills transfer straight to your own work.
- 💬 Sage Chat — the core streaming assistant, born in Section 6 when you first call an LLM from code.
- 📚 Sage Tutor — answers students' questions from Brightpath's own course docs, with citations, using Retrieval-Augmented Generation (Section 8).
- 🤖 Sage Research — a tool-using agent that plans, searches, and writes sourced reports, connected to real systems via MCP (Section 10).
- 🎨 Sage Studio — sees images, generates illustrations, and narrates lessons aloud (Section 12).
Prefer to build something else? Every project notes a swap. Sage Tutor's retrieval works just as well as a legal-document Q&A bot or a personal research assistant — same skills, your domain. Follow Brightpath for the worked examples, then point the final capstone at whatever you actually want to ship.
A layer per section
Here's the promise made concrete. Each section ends with a checkpoint — "Sage can now ___" — so you always feel the product getting more capable in your hands.
| Sections | What Sage gains |
|---|---|
| 2–4 · Foundations & LLMs | (You gain) an accurate mental model of how it all works |
| 5 · Prompting | A reliable personality and structured, parseable output |
| 6 · APIs | Sage Chat — a real streaming assistant |
| 7–8 · Embeddings & RAG | Sage Tutor — cited answers from Brightpath's docs |
| 9–10 · Frameworks & Agents | Sage Research — reasoning, tools, and MCP |
| 11–12 · Fine-tuning & Multimodal | A custom model and Sage Studio |
| 13–15 · Eval, Production, Design | Evaluated, secured, deployed, and scalable |
| 16–17 · Projects & Capstone | A portfolio and one integrated, shippable system |
Keep one Git repository for Sage from day one and commit at the end of every section. By the capstone you'll have a clean, narrated history of a production-shaped system — which, presented well, is exactly the portfolio piece that gets you hired (we'll cover that in Section 18).
What "done" looks like
The finish line is the capstone in Section 17: a single deployed application with user authentication, retrieval over private docs, an agent that calls tools, an evaluation suite that gates changes, monitoring, and a real deployment — plus an architecture diagram and documentation. That sounds like a lot today. By the time you arrive, you'll have built every piece at least once, and the capstone becomes assembly rather than invention.
Spend three minutes making this course personal:
- Of Sage's four faces — Chat, Tutor, Research, Studio — which would help your world the most, and why?
- Name a domain you'd swap Brightpath for in the capstone (your job, a hobby, a side project).
- Write one sentence: "By the end, I want to have built ___." Keep it somewhere you'll see it.
- You'll build Sage, an AI assistant for the fictional startup Brightpath.
- Four faces: Chat (§6), Tutor/RAG (§8), Research/agent (§10), Studio/multimodal (§12).
- Every section adds one concrete layer — checkpoints read "Sage can now ___."
- The capstone integrates everything; the alt-track lets you swap in your own domain.