Technical
The bits we wish we’d read three years ago. Ranking math, RAG failure modes, hybrid retrieval, evaluation done right.
A deep technical writeup on adding vector search capabilities to Chorus — the open-source search-relevance workbench — including architecture, indexing, and demo workflows.
How vector search reshapes e-commerce retrieval — semantic similarity, natural-language queries and the honest limits of pure dense retrieval.
A step-by-step guide to wiring Quepid — the search-relevance judgement tool — to a .NET search API for offline evaluation.
A practitioner’s look at using Ragas to evaluate retrieval-augmented generation pipelines — what it measures, what it misses, and how to combine it with judgement sets.
How to build a trustworthy evaluation harness for a RAG system in a week — judgments, offline metrics, LLM graders, and the traps that make numbers lie.
Embedding-based retrieval was supposed to make everything better. On real corpora with real queries, it often doesn’t — unless you invest in the four decisions that make hybrid work.
Discuss a project
If any of these deep-dives look like a problem you’re facing, that’s exactly the kind of engagement we take on.
