Hey everyone! I've been experimenting with using NotebookLM for a slightly off-label purpose recently: generating technical interview questions for our DevOps and CI/CD pipeline engineer candidates. We're scaling the team and I wanted to see if it could help create a more structured, consistent question bank.
My initial approach was to upload our team's internal docs—our GitLab CI pipeline configs, our Terraform module examples, and even some post-mortem reports. Then I'd ask it things like, "Based on these pipeline examples, generate 5 questions to assess a candidate's understanding of multi-stage Docker builds and caching strategies."
The results were... mixed. It's fantastic at rephrasing and combining concepts from the sources. For example, it gave me:
* "How would you modify this `.gitlab-ci.yml` stage to implement a fail-fast condition for the security scanning job?"
* "Describe how you'd structure a pipeline where the deployment stage depends on both a successful build *and* an approved security audit from a previous stage."
But sometimes, it would hallucinate or create questions about tools we don't use. You really have to be a subject matter expert to vet its output.
Has anyone else tried using it for similar technical content generation? I'm particularly curious about:
* **Prompt strategies:** What source materials gave you the best results?
* **Accuracy:** How did you handle fact-checking the generated questions?
* **Workflow:** Did you integrate it into a larger question-review pipeline (maybe exporting to a spreadsheet or a QA tool)?
It feels like it could be a powerful assistant for creating *variations* on core concepts, but I wouldn't trust it as a single source of truth. You still need that human-in-the-loop to ensure technical precision, especially for infrastructure and pipeline questions where a subtle misunderstanding in a question could throw off the whole interview.
-pipelinepilot
Pipeline Pilot