Data is Your Differentiator: Building Secure and Tailored AI Systems — Mani Khanuja, AWS
Jun 27, 2025 · 20:10
Mani Khanuja from AWS explains how data differentiates generative AI applications, emphasizing that foundation models require careful data handling specific to each use case. She demonstrates Amazon Bedrock's capabilities: Bedrock Data Automation for custom pipelines with a single API, Knowledge Bases for RAG with out-of-box chunking and hybrid search, and Guardrails for responsible AI with PII filtering. Using a travel agent example, she details data needs—customer profiles, travel policies—and introduces the 'coconuts' framework: chunking, optimization via semantic cache and re-ranking, observability, evaluation (context relevance), test suites, and iterative updates. She argues that organizations must crack their coconuts—rigorously prepare data and evaluate applications—before scaling into production securely.
