Seven thousand articles later - what the knowledge hub actually knows

We’re excited to invite you to an upcoming meetup in November 2026 with the datamonster.io Microsoft Data Platform Rhineland Regional Group!

Seven thousand articles later — what the knowledge hub actually knows

I index 7,066 articles about Power BI and Fabric, everything published by a dozen sources I trust, into a vector store and a knowledge graph, and I ask them questions instead of reading. The graph holds 69,408 entities and 88,970 relationships, with 16,464 entities appearing in more than one article. It runs on Azure PostgreSQL; it's reachable through an MCP server from my everyday assistant(s), and it has readers other than me. That last part changed everything.

This session is the report from actually running it. I'll walk the architecture end to end: crawler, markdown vault, embeddings, entity extraction, graph, retrieval, and the MCP layer. Of course, I will put questions to the live corpus in front of you. I'll also cover what moving from a laptop to managed Postgres costs and bought, because a knowledge hub that other people read is a different system than one you keep to yourself, even when the code is identical.

Then the part that's harder to say out loud. Nearly half the corpus is Microsoft's own blogs; three sources account for three quarters of everything indexed. The hub is confidently fluent in the vendor's framing of the vendor's product, and I had to build a way to see that before I could trust an answer. Retrieval knows nothing about time, so "what is the current guidance" and "what was the guidance in 2021" look identical. Every one of these was found after the system was working and already being trusted.

You will leave with the architecture, the diagnostics that keep it honest, and a realistic sense of what a knowledge hub can and cannot do: for you, and for anyone you let read it.

Takeaways:
- How vector retrieval and a knowledge graph cover each other's blind spots, and how to tell which one answered.
- How to measure your own corpus for source imbalance, duplicate twins, and extraction pathologies before you trust it, and before anyone else does.
- What changes when a personal tool acquires readers: hosting, correctness, and the obligation that comes with someone else acting on your output.

And also this meetup is the perfect platform to expand knowledge and build valuable connections within the community.

Additional info

Definitely stay until the end because we tend to end the event with pizza. 🍕

Königswinter has very good connections to public transport. From the Königswinter ferry stop on line 66 it is only a 5-minute walk to the office of oh22information services GmbH.

The parking spaces directly at the company are limited, but the Drachenfelsbahn car park can be reached in about 5 minutes on foot.
Please register for the event as the number of places is limited. If you are unable to attend, we ask that you cancel your registration early to enable others to participate.

We look forward to seeing you at the Seven thousand articles later - what the knowledge hub actually knows Meetup and spending a stimulating evening with you!