Library
Papers, articles and books I love and often refer back to.
Lost in the Middle: How Language Models Use Long Contexts ↗
Why retrieval that puts the right chunk in the prompt can still be ignored.
arXiv
Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks ↗
The paper that named the pattern, and the baseline everything since is measured against.
arXiv
Everything You Always Wanted to Know About Compiled and Vectorized Queries But Were Afraid to Ask ↗
The modern follow-up. Measures the two execution models against each other instead of arguing.
PVLDB
Billion-scale similarity search with GPUs ↗
The Faiss paper, and the reference point for approximate nearest neighbour indexes.
arXiv
Efficient and robust approximate nearest neighbor search using Hierarchical Navigable Small World graphs ↗
The graph index behind most vector databases, with the intuition for why it works.
arXiv
Choose Boring Technology ↗
Three innovation tokens per company, and the questions to ask before spending one.
mcfunley.com
The Tail at Scale ↗
Why one slow shard sets the latency of the whole request, and the tricks that hide it.
CACM
CAP Twelve Years Later: How the "Rules" Have Changed ↗
Brewer on consistency as a scope you choose, not a property of the system.
IEEE Computer
Dynamo: Amazon's Highly Available Key-value Store ↗
The paper that made quorum reads and eventual consistency practical at scale.
SOSP
Life Beyond Distributed Transactions: An Apostate's Opinion ↗
Helland's case for partitioning work by entity so you never need a distributed transaction.
CIDR
Architecture of a Database System ↗
The canonical tour of the layers, from process model down to query execution.
Foundations and Trends in Databases
Out of the Tar Pit ↗
The cleanest account of accidental complexity, and what to do about state.
Moseley and Marks
MonetDB/X100: Hyper-Pipelining Query Execution ↗
Where vectorized execution came from, and why it beat both interpreted and compiled plans.
CIDR
Time, Clocks, and the Ordering of Events in a Distributed System ↗
Where happens-before comes from, and still the clearest account of why clocks lie.
CACM