Reading

14 items · 1 in progress

Bookshelf

What I've worked through on the way to building backend and AI infrastructure — databases and distributed systems, then the papers behind modern inference.

Books

7

Read cover to cover unless marked otherwise.

Cover of Designing Data-Intensive Applications

Designing Data-Intensive Applications

Martin Kleppmann

Replication, partitioning and consensus — the spine of everything backend I've built since.

Cover of Build Your Own Database From Scratch in Go

Build Your Own Database From Scratch in Go

James Smith

B+tree to SQL in 3000 lines. The one that made storage engines stop feeling like magic.

Cover of Hands-On Large Language ModelsReading

Hands-On Large Language Models

Jay Alammar & Maarten Grootendorst

Read alongside my LLM-from-scratch work — tokenizers, attention, and generation end to end.

Cover of System Design Interview, Vol. 1

System Design Interview, Vol. 1

Alex Xu

The standard vocabulary for talking about scale.

Cover of System Design Interview, Vol. 2

System Design Interview, Vol. 2

Alex Xu & Sahn Lam

Harder, more specific systems — search, ads, payments, metrics.

Cover of Introduction to Algorithms

Introduction to Algorithms

Cormen, Leiserson, Rivest & Stein

CLRS, third edition. Referenced more often than read cover to cover.

Cover of Elements of Programming Interviews in Python

Elements of Programming Interviews in Python

Aziz, Lee & Prakash

Problem drilling, in Python.

Papers

7

The systems and inference papers I keep coming back to.

Cover of TAO: Facebook's Distributed Data Store for the Social Graph

TAO: Facebook's Distributed Data Store for the Social Graph

Bronson et al., USENIX ATC 2013

A graph API over sharded MySQL, and the read path that makes it work.

Cover of Scaling Memcache at Facebook

Scaling Memcache at Facebook

Nishtala et al., NSDI 2013

Cache invalidation, leases and thundering herds at a scale where every edge case happens.

Cover of FlashAttention

FlashAttention

Dao, Fu, Ermon, Rudra & Ré

IO-awareness as the actual bottleneck — tiling attention to stay in SRAM.

Cover of FlashAttention-2

FlashAttention-2

Tri Dao

Better work partitioning and parallelism over the original kernel.

Cover of Efficient Memory Management for LLM Serving with PagedAttention

Efficient Memory Management for LLM Serving with PagedAttention

Kwon et al. (vLLM)

Virtual memory paging applied to the KV cache. Directly shaped how I think about inference serving.

Cover of Mamba: Linear-Time Sequence Modeling with Selective State Spaces

Mamba: Linear-Time Sequence Modeling with Selective State Spaces

Gu & Dao

Selective state spaces as an alternative to quadratic attention.

Cover of AIOS: LLM Agent Operating System

AIOS: LLM Agent Operating System

Mei et al.

Scheduling and context management for agents — background reading for StacyVM.

GitHub
LinkedIn
X