Hey everyone! I've been experimenting with NotebookLM for the past few months, mainly for internal team training and documentation. I love how it can ground answers in our own source material, but I keep hitting a wall when it comes to connecting it with our actual cloud infrastructure.
My team's SaaS platform lives on AWS, and we have a ton of operational docs, architecture diagrams (in text form), and support logs in Google Drive. The dream workflow would be to have NotebookLM analyze our AWS cost reports or CloudWatch log summaries that we store in Drive, and then ask it questions to optimize spending or troubleshoot common errors. But I'm stuck in a manual loop of uploading PDF exports.
So, my big question: has anyone actually managed to integrate NotebookLM with live or semi-live data from AWS or GCP? I'm not talking about the basic Google Drive connection for static docs, but something more dynamic.
For example, are you using a pipeline to dump daily billing exports or BigQuery summaries into a Drive folder that NotebookLM then indexes automatically? Or have you found a way to use the API to feed it structured data from these platforms? I'm particularly curious about real deployment stories—what was the use case, how did you set it up technically (even at a high level), and did the benefits outweigh the setup hassle?
I'm worried about data security too. Feeding sensitive cloud metadata into a third-party tool gives me pause, so any insights on how you've handled that would be incredibly helpful.
You're right to focus on the operational data gap. The native Google Drive integration is for static documents, not a data pipeline.
We've established a semi-live workflow for cost analysis. Each morning, an AWS Glue job transforms the CUR into a simplified CSV summary, which is automatically placed into a dedicated Drive folder. NotebookLM indexes it by noon. It works for high-level Q&A on service spend trends, but the latency means it's useless for real-time troubleshooting.
The critical caveat is data structure. NotebookLM struggles with raw, nested JSON from CloudWatch logs unless you pre-process it heavily into plain text narratives. For log analysis, we've had better results with a dedicated internal tool that uses the API directly.
Trust but verify. Then renegotiate.