How do we check that generated prose contains the right information, rather than merely sounding like a good answer? Ask an agent for a refund status when the authoritative tool says pending. “Your refund is complete” answers the question but contradicts the…
In the first StyloMail article, I used Jev to ask specific questions about email and get typed, probabilistic answers back. Nimble, from Bespoke Labs, brings that technique to a model I can host myself through Ollama. From C#, I can send the same kind of state…
A Ten Commandments–Compliant Architecture In which we build a voice-to-form system that doesn't hand the keys to the AI kingdom Here's the thing about voice interfaces: everyone wants them, nobody trusts them. And they're right not to. Most voice forms fail…
Small and local LLMs are often framed as the cheap alternative to frontier models. That framing is wrong. They are not a degraded version of the same thing. They are a different architectural choice, selected for control, predictability, and survivable failure…
GitHub release .NET Version This is Part 2 of the DocSummarizer series. See Part 1 for the architecture and patterns, or Part 3 for the deep technical dive into embeddings and retrieval. Turn documents or URLs into evidence-grounded summaries - for humans or…
Here's the mistake everyone makes with document summarization: they extract the text and send as much as fits to an LLM. The LLM does its best with whatever landed in context, structure gets flattened, and the summary gets increasingly generic as documents get…
Series: Local LLMs for Data - Part 1 of 2 Here's the mistake everyone makes: they try to feed their CSV into an LLM. Don't. LLMs should generate queries, not consume data. You've got a 500MB CSV file and want to ask "What's the average order value by region?"…