A personalized learning assistant on AWS rarely fails loudly. It fails by being helpful to the wrong student. Five failure families that decide whether an AI tutor built on Amazon Bedrock is safe to put in front of a class: LTI identity, retrieval isolation, guardrail limits, the cost shape of an academic term, and being able to prove what the assistant saw.
Pool timeouts while RDS sits at low CPU? The database isn't the bottleneck. Here's how to tell whether your app pool, max_connections or RDS Proxy ran out, the SQL that proves it, and the fixes that survive the next deploy.
The assessment report grades syntax, not behavior. A four-tier breakdown of what AWS SCT converts cleanly, what it converts while silently changing meaning, what it only handles through the extension pack, what it refuses outright, and the whole category it never looks at.
Latency charts don't decide this. The query nobody wrote down at the start does. How to build an access pattern inventory, where DynamoDB and RDS each genuinely win and hurt, a six-question decision procedure, and the arguments on both sides that fall apart under a real workload.
A missing tenant filter doesn't throw an error, it returns a 200 with too many rows. This is how to build a multi-tenant workshop management platform on AWS where the isolation boundary sits below your application code: STS session tags feeding IAM conditions, DynamoDB leading keys, scoped S3 prefixes, forced PostgreSQL row-level security, and a control plane that verifies each new tenant is fenced before anyone logs in.
Most Oracle to Aurora migrations don't fail loudly. They finish green, pass validation, and quietly hand you truncated LOBs, sequences stuck at 1, and empty strings that used to be NULL. Here are the four failure families to check before you cut over, with the SQL and task settings that catch them.