Writing · 13 posts

Notes on AI, backend engineering
and the bits worth remembering.

LLMs in real systems, backend fundamentals, debugging stories, and things I tried that may or may not have worked.

Databases 2026-10-01

ACID in practice: what SQLite, Postgres, MySQL and MariaDB actually promise

Atomicity, consistency, isolation and durability, one letter at a time, and how differently SQLite, Postgres, MySQL, MariaDB and a few others implement each one. Same words, different guarantees.

Networking 2026-09-26

TCP: how a connection actually works

What TCP does on top of IP: ports, the three-way handshake, sequence numbers, ACKs, retransmission, closing a connection, and the costs that pushed HTTP/3 onto QUIC.

NLP 2026-09-25

Vectorizing text: from word counts to meaning

How text becomes numbers a model can work with: one-hot vectors, bag of words, TF-IDF, the hashing trick, word embeddings and sentence embeddings, and when to use each.

Networking 2026-09-16

The OSI model, layer by layer

What each of the seven OSI layers does, how one HTTPS request gets wrapped and unwrapped on the way to a server, and why switches, routers and load balancers stop at different layers.

LLM 2026-09-12

Five steps from a business problem to an LLM that actually ships

Understand, prepare, select, customize, productionize. A working order of operations for picking a language model and getting it into production without burning a quarter.

NLP 2026-09-11