The journal
Essays with runnable code and interactive demos — from React frontends to JVM backends, data pipelines to LLM-powered features. Written by a fullstack developer working in AI. Browse by category or tag.
Retrieval-Augmented Generation grounds an LLM's answers in your own documents. Here's the whole idea in one diagram, then a complete working system in ~80 lines of Python — core pipeline executed and verified.
Naive RAG fails silently — the answer looks fine while the retrieval missed. Here's the production upgrade path: recursive chunking, hybrid search with RRF fusion, cross-encoder reranking, query rewriting, and an eval harness that tells you which stage to fix. Core algorithms executed and verified.
Token pricing looks simple until you're running a million calls a month. Here's a practical way to count the real cost — and a small harness for trading cost against quality.
Your code runs top to bottom. Spark doesn't. A small-dataset walkthrough of lazy execution order, shuffles, caching, persist, checkpointing — and the best practices that fall out of them.
JDK 27 shipped September 15, 2026 with 9 JEPs. Four are final — and two of them change how every JVM you run behaves without touching a line of code.
NVIDIA's NOOA framework treats an AI agent as a plain Python object: fields are state, methods are capabilities, docstrings are prompts, and an ellipsis body means 'the LLM implements this.' Here's how it works and how to build your first agent.
