System Design 301: AI, agents and data
The app has grown and starts using AI: a support chatbot, an agent that handles complaints, reports for thousands of restaurants. You learn to design these parts so they run reliably, safely and at the right cost, and finish with a design project reviewed by an instructor.
- Advanced
- About 12 to 15 hours, plus the project
- 8 sections
- 51 lessons
This course is being written and is not open yet. The outline is below; leave your email if you'd like to hear when it opens.
Tell me when it opens
Course outline
Section 1
Getting started
Where are you now?
Check · 10 min
When a request costs money per word
Lesson · 10 min
Section 2
The customer support chatbot (LLMs in production)
Every answer has a price
Lesson · 10 min
The answer shows up only after a long pause
Lesson · 10 min
Hitting the provider's limit
Lesson · 10 min
The same long instructions every time
Lesson · 10 min
"How much is delivery?" asked a hundred ways
Lesson · 10 min
The provider has an outage
Lesson · 10 min
The chatbot meets real customers
Practice · 25 min
Looking back at chapter 1
Lesson · 5 min
Section 3
Answering by the actual policy (RAG)
The chatbot invents a refund policy
Lesson · 10 min
Reading the wrong clause
Lesson · 10 min
The policy changed this morning
Lesson · 10 min
The answer must have the right shape
Lesson · 10 min
A test set that stops a bad release
Lesson · 8 min
Ask anything, answered correctly
Practice · 25 min
Looking back at chapter 2
Lesson · 5 min
Section 4
Which model for which job (AI costs)
Not every question needs the strongest model
Lesson · 10 min
Reading every customer review, every day
Lesson · 10 min
Night jobs need no instant answer
Lesson · 10 min
Run your own model or pay per token?
Lesson · 10 min
One user burns the whole budget
Lesson · 8 min
Answers change tone overnight
Lesson · 8 min
Weighing the AI bill
Practice · 25 min
Looking back at chapter 3
Lesson · 5 min
Section 5
An agent that handles complaints (agent loops)
An agent is a loop that costs money
Lesson · 10 min
Every step can be a little wrong
Lesson · 10 min
Context grows with every step
Lesson · 10 min
A tool never answers
Lesson · 10 min
The server restarts mid-task
Lesson · 10 min
Complaints pile up after an incident
Lesson · 10 min
The agent's first shift
Practice · 25 min
Looking back at chapter 4
Lesson · 5 min
Section 6
The agent must not refund at will (agent safety)
Check before the agent acts
Lesson · 10 min
A review that gives the agent orders
Lesson · 8 min
Only the access it needs
Lesson · 8 min
Which refunds may the agent decide alone?
Lesson · 10 min
Several layers of protection
Lesson · 8 min
A day full of complaints
Practice · 25 min
Looking back at chapter 5
Lesson · 5 min
Section 7
Reports for thousands of restaurants (data platform)
Reports slow down ordering
Lesson · 10 min
This month's report reads everything since launch
Lesson · 10 min
Data trickles into the warehouse
Lesson · 10 min
Last night's report ran until morning
Lesson · 10 min
Revenue higher than the money received
Lesson · 10 min
The source changes its shape
Lesson · 10 min
Reports ready before opening time
Practice · 25 min
Looking back at chapter 6
Lesson · 5 min
Section 8
Capstone project
The brief and the rubric
Lesson · 20 min
Build and test your design
Practice · 60 min
Your design and its reasoning
Project · 120 min
How far have you come?
Check · 10 min
How will you design your next system?
Lesson · 5 min
Linked courses
Suggestions, never requirements: every course is open on its own.