In this video, I build what I call a “Memory Gem” — a Gemini Gem that actually knows your business because it’s grounded in a living memory document. The key idea is simple: anything I add to that memory document automatically updates the Gem, so my assistant keeps getting smarter without me rebuilding prompts every time. I show you how I pull data from real sources like Google Drive files, meeting transcripts, and anything I can load into NotebookLM, then refine and expand it into a structured set of headings that Gemini can use properly. I also demonstrate one of my favourite NotebookLM features: Discover. Once you’ve added a few sources, you can click a button and NotebookLM will suggest other relevant files in Drive to round out the context (sometimes it finds weird stuff — you’ll see that happen). From there, I paste new info into the memory document with a clear instruction: ignore duplicates, keep anything additive, and don’t remove existing context. Finally, I test real-time updates by adding team details from leadership meetings and immediately asking the Gem what it knows about a person — and it pulls the new info straight away. The takeaway: iterate bit-by-bit, and your Gem becomes “10 times better” as your business context grows.