# From broken preview connectors to a working local RAG Hangman word selector, what I learned fighting Semantic Kernel 1.70.0-preview

Spent way too many hours trying to get a simple local vector store working with Semantic Kernel previews for AI-102 prep. I tried 2 different Connectors and both were not working!

InMemoryVectorStore → "collection does not exist" even after GetCollection + dummy upsert
SqliteVectorStore → NOT NULL constraint on json column

Ended up with a manual raw-SQL upsert workaround + code-level category filtering + random pick from top-4 results. Query: "fruit food category" → Top=4 → pick one randomly → Hangman word ready

Lesson: previews are rough, but manual control + persistence beats fighting abstractions. Exam on the 30th — wish me luck! Code is already pushed to my repo: [SKOrchestrationPractice](https://github.com/P47K0/SKOrchestrationPractice)

![Vector store word-selection code screenshot](https://images.koorevaar.com/linkedin-import/hangman-rag-vector-store-troubleshooting.jpg)

---

*Co-authored with Claude.*

If this was useful, [you can buy me a coffee](https://ko-fi.com/p47k0).

New articles: [follow via RSS](https://blog.koorevaar.com/feed.xml).