Technical explorations of information retrieval, safe autonomous agents, vectorized database internals, modern web architectures, and LLM evaluation frameworks.
Latest, , 7 min read
Pure vector search frequently fails on exact keyword identifiers and domain jargon, while lexical BM25 misses semantic paraphrasing. This article breaks down the mathematical mechanics and implementation of Reciprocal Rank Fusion (RRF) for production RAG pipelines.
RAG, Information Retrieval, BM25, Vector Search, Python
7 min read
Giving language models unbounded code execution in analytical pipelines opens serious security and hallucination risks. This guide shows how fixed tools, SQL restrictions (keyword filtering versus AST parsing), an in-process DuckDB connection and Pydantic schema contracts narrow what an agent can do, and where those controls stop short.
AI Security, DuckDB, Python, Pydantic, SQL
5 min read
Moving analytical data out of a database server and into the application process removes network and serialization overhead for local workloads. This article explains how DuckDB combines columnar storage, vectorized execution and morsel-driven parallelism, where it fits, and where its limits are.
DuckDB, OLAP, Database Internals, Data Engineering, Performance
6 min read
React Server Components change where code runs and what is sent to the browser. This article explains how the Next.js App Router uses Server and Client Components, what can cross the server-client boundary, how Suspense streaming works, and how Incremental Static Regeneration revalidates pages.
Next.js, React, TypeScript, Web Architecture, Performance
5 min read
A basic RAG pipeline is quick to build; measuring how faithful its answers are is harder. This article explains common RAG evaluation metrics, what a simple citation check can and cannot show, and how to keep most tests offline and cheap.
AI Evaluation, RAG, Hallucination, Testing, Python