The correct option is Implement retrieval-augmented generation against the company's secured document repository. This keeps the model grounded in your internal briefs and case summaries and ensures responses are constrained to those sources rather than the open web or the model's general knowledge.

With RAG you index the past 36 months of documents and retrieve only the most relevant passages at query time which are then provided to the model as context. You can apply enterprise access controls and include citations so the assistant can answer only from approved internal materials and provide provenance. In Vertex AI this pattern is supported through grounding with enterprise data using services such as Vertex AI Search or Agent Builder so the model remains constrained to your repository.

Enable Vertex AI Search with public web connectors to broaden coverage is not appropriate because bringing in public web data contradicts the need to answer strictly from internal documents and would reduce control over source provenance.

Connect the assistant to a live internet search API for up-to-date answers would surface external web content which conflicts with the requirement to avoid public information and general model knowledge in responses.

Fine-tune a foundation model on publicly available industry whitepapers introduces public content that you explicitly want to exclude and fine-tuning does not guarantee the model will limit answers to your internal briefs. Fine-tuning modifies weights rather than enforcing strict source grounding and can still produce unsupported or generalized outputs.

Full AWS Practitioner Certification Question