Pick a depth. Each prompt opens in your AI pre-loaded with the lesson. Click a row to preview the prompt.
Build order is a design decision, not a scheduling one. The first project gives you a system worth evaluating; the evaluation project measures it; the red-team project attacks it; the distillation project makes it cheaper. When your repositories reference each other, the portfolio reads as an engineer who built a system. When they are six unrelated demos, it reads as someone who worked through a list. This matters more than adding a seventh project, because the follow-up question in an interview is almost always 'how do these relate' — and a dependency chain answers it for you.
The dependency chain below is the sequence used in this course. If you only have time for two, build the retrieval project and the gateway: retrieval plus infrastructure is the most common shape of a real AI engineering job.
# The chain — each arrow is a real dependency, not a suggestion.
#
# Project 1 Knowledge-graph RAG ........ the system under test
# |
# +--> Project 4 Judge ............ measures Project 1's answers
# | |
# | +--> Project 7 Regression CI ... gates changes on that judge
# |
# +--> Project 6 Red team ......... attacks Project 1 and 2
# |
# +--> Project 5 Distillation ..... makes Project 1's calls cheaper
#
# Project 2 Research agent ............. second target for the red team
# Project 3 Gateway ................... every other project calls through it
#
# Practical consequence: build 1 first, and route 2-6 through 3 once it exists.
# Your gateway then has real traffic from your own systems, which is far more
# convincing than synthetic load.python3 main.py