Starting from building a knowledge base for my thesis system, this guide sets out when knowledge-base AI projects are a good fit for NotebookLM.
- Starting point: building an “AI parametric furniture support bot” for my thesis — with a GPT knowledge base, data structures were hard to feed, image–text links were read poorly, and iteration was costly
- Pairing NotebookLM with Gemini turned PDFs of parameters and spec notes into a knowledge base you can chat with and trace back to sources — get results first, then decide whether to engineer it
- Four project types suited to adoption: document-heavy support or expert systems, research and design projects needing frequent image–text comparison, solo developers who don’t want to write RAG first, and internal tools whose documents change often — each with an adoption approach
- A five-question checklist (large body of existing documents? knowledge-centered? need traceable answers? want a PoC first? is the bottleneck in organizing data?): three or more “yes” answers mean it’s worth adopting