Thoughts on private AI, data sovereignty, and building in the open.
An agent-ready design system is half readable, half enforceable. We tested five claims ours made about our code, all five were wrong, and rebuilt it.
In August we removed our entire agent development framework. What the research says about context files, why modern coding models need skills less, and the five layers we kept.
We found 56 hardcoded colour values our own design system forbids. What the CHI 2026 research says about AI agents and design systems, and what we built.
A plain-language explainer of large language models. What they are, how they answer questions, where the math happens, and what to ask any AI vendor about your data.
Klai is becoming a steward-owned company. What steward ownership means, why a not-for-sale structure fits our mission, and how we're getting there.
A PDF, a meeting transcript, and a knowledge base article should not be processed the same way. How content profiles control chunking, context, and question generation.
Follow a single user question through a retrieval pipeline: from pronoun resolution to three parallel searches, rank fusion, and a final reranker pass. Discover why one similarity score is not enough.
AI needs a foundation of compliance and digital sovereignty. That's why we're building Klai from Groningen, for organisations that won't hand over their data.
Reduce latency, avoid irrelevant answers, and save GPU costs by skipping knowledge base retrieval for queries that are not questions. How pattern matching and semantic gates filter trivial queries before they hit the vector store.
Know if your knowledge base fix worked with automatic gap re-scoring. Learn how retrieval validation closes the loop between edits and outcomes.
Turn failed searches into a content roadmap. Learn gap detection methods, hard vs soft gaps, and how to systematically fix knowledge base coverage.
How to bridge the gap between user questions and knowledge base answers. Techniques that work at storage vs query time, and why the retrieval system isn't where the fix happens.
Retrieval tuning can't fix a storage problem. Learn why knowledge base quality is determined at ingest time, and what contextual retrieval and HyPE actually do.
AI can classify documents reliably. It cannot design taxonomy structure. Learn why every production knowledge base needs human governance for categories.
Why single type fields fail in knowledge systems. Learn how provenance origin, assertion mode, and synthesis depth drive better metadata design for knowledge bases.
Evidence vs claims: why source documents and knowledge artifacts require different handling in knowledge systems. A critical distinction for organizational knowledge.
Why decision data models matter: structure your organizational decisions with 5 fields. Capture rationale, alternatives, and consequences to build organizational memory.
How to organize Claude Code rules and CLAUDE.md into layers so the right context loads when you need it, and every mistake improves the system permanently.
Knowledge bases fail when they compete with culture. Handbook-first organisations document by default, and that foundation is what makes AI-augmented knowledge tools actually useful.
Most knowledge bases decay silently. A self-improving system detects gaps, prioritises fixes, and stays accurate without constant curation. Here is how.
Most companies delete outdated knowledge. That destroys institutional memory. Here is why knowledge retention means evolving, not overwriting.
Most organizational knowledge lives in heads, Slack threads, and meetings. Here is why tacit knowledge loss is expensive, and how to fix it.
EU AI Act compliance is now live. Here is what it means for law firms, financial advisors, and healthcare providers using AI day to day.
Most AI tools send your data to US servers. Here is what that means for GDPR compliance, and what your privacy-first options are.