Standard RAG Should Be a Tool, Not the Architecture
Why one-shot retrieval breaks on ambiguous and multi-step questions, and why standard RAG works better as a fast retrieval tool inside an adaptive agentic architecture.
Read it →Thinkthen exist
A humble notebook — what I think, what I test, what I build, what breaks, and what I make of it.
Notebook
Experience
Field notes from shipping systems — what held up, what did not, and what I would do again.
Latest: How I Make One Page Citable, and How I Check ItExperiments
Labs, prototypes and unfinished tests. Useful even when they fail.
Latest: Can Jev Make Large-Scale Chat History Analysis Practical?Ideas
Essays and arguments that are not a product yet.
Latest: Standard RAG Should Be a Tool, Not the ArchitectureGPU & inference
Serving, kernels, quantization and the cost of actually running the model.
Nothing in this section yet.All writing
01
RAG
Standard RAG Should Be a Tool, Not the Architecture
Why one-shot retrieval breaks on ambiguous and multi-step questions, and why standard RAG works better as a fast retrieval tool inside an adaptive agentic architecture.
02
GEO
GEO Is Not a Ranking. It Is Three Events You Can Measure.
Search rankings do not tell you whether an answer engine fetched your page, used it, or named you. Generative engine optimization starts by keeping those three events apart, and by refusing to treat a benchmark lift as a promise.
03
GEO
How I Make One Page Citable, and How I Check It
A citable page answers one question, puts the source next to the claim, and can be checked without pretending an unmeasured signal is a pass. Here is the method I actually use.
04
Jev
Can Jev Make Large-Scale Chat History Analysis Practical?
Jev is a newly released decision model from TypeSafeAI. I test whether its combination of seconds-scale classification, very low inference cost and usable semantic accuracy can make large-scale chat-history analysis practical.
05
RAG
Why AI Projects Die Between the Demo and Production
Confusion around enterprise AI integration is no longer unusual. Executives develop high expectations, a technical team produces an impressive demo within weeks, early users see the potential—and then the project gets close to production. Costs rise, latency appears, security questions surface, retrieval quality becomes inconsistent, and an architecture that worked for a handful of developers behaves very differently when hundreds of employees arrive.
06
RAG
Document Processing for Enterprise RAG
A company rarely has one document to connect to an AI system. It has hundreds of policies, contracts, scanned PDFs, technical reports, presentations, tables and Excel workbooks. From the outside, the task looks straightforward: upload the file, extract the text, split it into chunks and send it to a vector database.
07
EnterpriseAI
Enterprise AI Integration: From Prototype to Production
Over the past few years, I have seen many companies struggle with the same uncertainty around AI integration. Almost everyone agrees that AI represents a major transformation. Bringing it into business processes at the right time is now widely treated as a strategic priority.
Contact us
A question, a project, or something that should not stay in a comment box.