Latest 6 min read
How LLMs actually work
Tokens, embeddings, attention and next-token prediction — what the machine is really doing, why it hallucinates, and what 'reasoning' is and isn't.
3 posts
Writing.
Notes on building and shipping production web software — architecture, tooling, and the things that only show up once real users arrive.
- / 7 min read
How to use a vector database
Embeddings, similarity search and RAG — how retrieval actually works, why chunking and hybrid search decide your quality, and when a vector database is the wrong tool entirely.
- / 7 min read
How to retain context
Context windows are the working memory of an LLM application, and they fill up faster than anyone expects. What actually consumes them, why long conversations degrade, and the techniques that work.