Quotes all the way down
After building RAG systems for technical books, I kept asking the same question. Which chunking strategy actually retrieves better? I built a research platform to find out, with real vector search, LLM-as-judge ground truth, and proper IR metrics.
How I work through technical textbooks with an LLM as a study partner, chapter by chapter, with room for side quests and real code along the way.
Complect gets arrays, a 1993 demoscene fire effect, and a VS Code language server. Building all three revealed exactly what the language still needs.
Complect grew from a Babel-based transpiler into a multi-backend compiler with an LLVM code generator, user-defined functions, manual memory management, and SDL graphics. Here is the architecture story behind that transformation.
Imagine AI whispering flawless code into your project, elegant and efficient. But what if that magic hides shallow understanding? My journey from sci-fi novels to real-world coding reveals how AI should amplify your skills, not replace the craft.
Imagine 1983, hammering commands into a CRT terminal, green glow burning your retinas as a sentient machine hums back. Now crank it to 2025, and I am unleashing HAL, a feral AI assistant that is half WarGames fever dream, half tech apocalypse.
Ever wondered if creating a compiler in Node.js is a good idea? I found myself pondering this very question while on the lookout for a side project to explore the depths of Node.js Streams.
Contributing to the Node.js project for the first time and understanding the processes around a large open-source project. Working with and learning from a welcoming group of people.