Ólafur Aron Jóhannsson

October 15, 2025 • 7 min read

What are Vector Embeddings?

Understanding vector embeddings: how machines turn words into numbers that capture meaning. Interactive demo included.

machine-learning embeddings bert nlp rust webassembly semantic-search

October 17, 2025 • 7 min read

BM25 vs TF-IDF: Keyword Search Explained

Understanding BM25 and TF-IDF: the keyword search algorithms that power everything from Google to your local search bar. Interactive comparison included.

search information-retrieval bm25 tfidf rust webassembly nlp

March 10, 2026 • 7 min read

Benchmarking Network I/O: From a Blocking C Server to a C++20 Framework

Blocking, thread-per-connection, single-thread epoll, multi-thread epoll. Same machine, same benchmark, same response. From 143K to 1.45M requests per second. Then a C++20 framework that matches the raw C.

C++ C epoll threads coroutines performance benchmarking networking lundi

March 11, 2026 • 10 min read

How Async I/O Actually Works: From read() to C++ co_await

A deep dive into async I/O in C and C++. Build an event loop from scratch, understand epoll and io_uring, and see how C++20 coroutines and co_await work under the hood.

C++ C async-io coroutines epoll io_uring event-loop concurrency

March 13, 2026 • 10 min read

Why std::string is Slow and std::string_view is Fast

Benchmarking std::string vs string_view (22x), vector vs list (12x), the real cost of std::unordered_map, and std::optional for lazy initialization. Real numbers on real hardware.

C++ performance string_view unordered_map benchmarking optimization containers cache-locality vector-vs-list

Blog

What are Vector Embeddings?

BM25 vs TF-IDF: Keyword Search Explained

Benchmarking Network I/O: From a Blocking C Server to a C++20 Framework

How Async I/O Actually Works: From read() to C++ co_await

Why std::string is Slow and std::string_view is Fast