Posts in "AI-engineering"

A technical analysis comparing how five different AI coding CLIs manage token usage and context windows when using the same language model to solve an identical debugging task.

Great breakdown of the hidden costs behind AI coding assistants. The differences in tool definition overhead alone are wild — Claude Code sends 62,600 characters of tool definitions per turn while Aider sends zero. Worth reading if you’re thinking about which CLI tools you’re reaching for and what you’re actually paying for under the hood.

Embeddable RAG library for Elixir/Phoenix with agentic pipelines and dashboard.

This looks like a really well-thought-out approach to adding RAG capabilities to Phoenix apps. The three-function API (ingest, search, ask) keeps things simple, but it also supports more advanced agentic pipelines with query expansion, multi-hop reasoning, and self-correction, with built-in LiveDashboard integration and swappable backends (pgvector, Bumblebee for local embeddings)