Participated in public demo 2603 with CBS intelligence layer

The problem [00:50] Real estate brokers, journalists, and analysts working with AI all faced the same wall: CBS (the Dutch national statistics agency) publishes vast, authoritative public data, over 5,900 tables and 150,000+ columns, but it's cryptically coded, inconsistently leveled across geographies, and entirely in Dutch, so asking an LLM directly about it produced confident, plausible-sounding answers built on hallucinated codes and invented numbers. What they built [01:25] He built the CBS Intelligence Layer: an MCP (Model Context Protocol) server that normalizes, indexes, and pre-computes CBS statistics for sub-second, source-traceable lookups (with live fallback to the full CBS catalog when something isn't pre-stored), paired with domain-specific Claude Skills that layer interpretation on top, a broker lens and a journalist lens, each producing a different expert reading of the same underlying facts while a shared trust discipline forces every claim back to its CBS source and refuses fabricated causal claims. What happened [03:50] Live runs on the same Amsterdam neighborhood (the Jordaan) showed the architecture holding up under two very different lenses, the broker skill returned a defensible, source-cited buyer's brief in about 30 seconds of compute (work that would normally take a junior broker hours), while the journalist skill surfaced a socioeconomic paradox in the same data (a 78-percentile gap between education and wealth) that the broker's reading had missed entirely; the broker skill took about a week to build, the journalist skill about a day, with 1.3M+ pre-stored CBS datapoints underneath both. The ask [07:30] Jacco is looking to connect with people interested in applying this reusable trustworthy-AI pattern, a shared data layer with swappable expert skills, beyond CBS, to any domain where trusted data and human expertise need to work together; reach out via the LinkedIn link on his closing slide.