Why Memory Matters

Concept. Databases give applications durable memory: data survives after a computation ends.

Intuition. Algorithms and models compute. Databases preserve what those computations read and write. Here, memory means persistent storage. RAM remains volatile working memory.

What this course teaches

Five module panels move from SQL and storage on one machine through indexing, transactions, and distributed systems across many machines. They lead into M6, which applies those mechanisms to phones, AI-agent memory, and modern data platforms.

The modules and the real systems each one explains. Along the way the figure names what each move buys: disk pages, a billion rows answered in milliseconds, all-or-nothing; declarative, on disk, many machines.
M1 · SQL	M2 · Storage	M3 · Indexing	M4 · Transactions	M5 · Distributed
Postgres · BigQuery	Chrome · Uber	Spotify · OpenAI	Stripe · GitHub	Netflix · Kafka

Figure 1. Each module adds one capability, and each one supports the next scale.

M1 shows you how to use a declarative, parallel database
M2 stores the rows on disk
M3 locates a row in milliseconds
M4 keeps it correct while thousands of users write at once
M5 runs one logical database across many machines
M6 applies these mechanisms to phones, AI-agent memory, and modern data platforms

The career map connects these modules to the projects and the paths they lead to. For what the course actually answers, the twenty questions and the level of the stack each is answered at are on the Modules page.

Where we start: three systems, reconstructed

We learn how the database is built by taking three real systems apart and finding the same primitive inside each one. In teaching order, from a laptop to the planet:

Three case studies progress from one machine to many. Small analytics compares Pandas in memory with Polars streaming. Agent memory separates working context from persistent memory. UberEats coordinates shared transactional state and saves phone actions through network failures.

The three opening teardowns. 0.1 Small analytics: Pandas and Polars on a laptop, fast until it runs out of RAM. 0.2 Agent memory: no memory, then local, then global. 0.3 A big service: UberEats, one database, three sides, plus the edge. The first row is the scale each one reaches, the second what it is here to teach. The arc runs from one machine to many machines.
0.1 · Small analytics	0.2 · Agent memory	0.3 · A big service
~16 GB on a laptop	1 user → 800 M users	billions of orders, planet-wide
the on-ramp	the whole course in one example	planet scale and the edge

Figure 2. Three case studies move from working state on one machine to persistent state across many machines. Modules 1 through 6 explain the mechanisms underneath.

0.1 Pandas: Deconstructing Small Analytics. Pandas starts in memory. Polars extends the range of one machine with lazy and streaming execution. What changes when the workload needs more than one machine?

0.2 Claude & OpenAI: Deconstructing Agent Memory. How does an agent preserve decisions outside its context window and retrieve them for a later request? What changes between local memory and a global product?

0.3 UberEats: Deconstructing a Big Service. How does a three-sided service coordinate shared state at scale? How does its driver app keep working through a network interruption?

Across all three cases, volatile state supports the current computation. Persistent storage carries data across processes, sessions, devices, and machines.

By the end of this course, you can design one. Start with 0.1 Pandas: Deconstructing Small Analytics.

Why Memory Matters

What this course teaches

Where we start: three systems, reconstructed

Video