// blog
Notes from the field.
Thoughts on data engineering, AI platforms, and lessons learned from real enterprise projects.
September 20, 2026
I Finally Built the Golden Dataset
Six articles ended with some version of “measure it.” This one shows how, starting with the dataset I spent most of the series admitting I didn’t have.
read
September 13, 2026
The Chunk Was Never the Document
Five articles of retrieval tuning assumed a chunk is a span of text you can embed, cite, and delete. Point that assumption at a scanned invoice or a 400 page equipment manual and watch it come apart. This is what replaces it, and what the replacement costs.
read
September 6, 2026
The Loop That Doesn't Know When to Stop
Adaptive RAG makes one decision before retrieval. Agentic RAG makes a decision after every step, and keeps making them until something tells it to stop. The loop was never the hard part. The stopping condition is.
read
September 1, 2026
GraphRAG in Practice
Part 1 said most GraphRAG builds are expensive ways to store the same chunks in a worse format. This is the argument in full, plus the cases where a graph is the only thing that works. The hard parts are all at index time.
read
August 25, 2026
When Your RAG Knows It's Wrong
Part 3 of the RAG Architecture Series. Corrective RAG and Self-RAG are two different answers to the same question: can the pipeline notice its own failure? One changes the pipeline. The other changes the model. Here's how they behave once real traffic hits them.
read
August 23, 2026
Late Interaction, SPLADE, and the Architectures I Still Do Not Ship
Addendum to Part 2. The four families between a bi-encoder and a cross encoder, what each one would have to prove, and the number that keeps deciding it for me.
read
August 16, 2026
Hybrid Retrieval and Reranking: The Part Where We Actually Build It
Part 2 of the RAG Architecture Series. Chunking, BM25 analyzers, fusion weights, cross encoders, and the mechanism behind the part number example I left dangling last time.
read
August 4, 2026
Which RAG architecture should you choose?
A 2026 decision guide for the people who sign off on the architecture and own the budget, not the ones writing the retrieval loop.
read