Event-driven AWS, in production
Ex-Amazon senior backend engineer, 15 years. I review EventBridge, SQS, Step Functions, Lambda and ECS stacks, including the LLM calls now running through them.
F-06 No DLQ on any target. A failed delivery is dropped.
F-05 Standard where Express fits: ~USD 1,950/mo vs ~USD 30 at 10 TPS.
F-03 Returns a failed PutEvents as success. Events vanish.
F-04 Queue policy accepts SendMessage from any AWS account.
Fixed-price audit
An independent senior review of your event-driven stack. What will break, what is costing you money, and the order to fix it in.
USD 3,500
Fixed price. 10 business days from kickoff.
First three audits booked before 30 November 2026: USD 2,100 in exchange for a named case study.
Small teams with coding agents now ship more infrastructure than anyone reviews. These are six of the areas every audit covers, and each finding comes with evidence from your account, not a checklist tick.
EventBridge rules, SNS and SQS fan-out, ordering, duplicates, idempotency.
Retries, backoff, DLQs and whether anyone reads them, poison messages.
Standard vs Express, Lambda sizing, log ingestion, NAT and data transfer.
Alarms that would actually page, tracing coverage, structured logs.
Over-broad roles, resource policies, secrets, public endpoints.
Model calls inside queues and workflows: retries that re-bill tokens, provider throttling, timeouts, spend nobody alarms on.
From the public sample audit of a published AWS reference application, reviewed exactly as a client stack would be.
PutEvents is reported as successThe function that publishes every downstream event returns an error object instead of throwing. The workflow ends green, nothing pages because there are no alarms, and the events are gone.
Fix: throw on FailedEntryCount > 0. One line, same afternoon.
ForAllValues:ArnEquals on a single-valued key passes when the key is missing.
A two-state Standard workflow on the per-transaction path. Express runs it for ~USD 30.
Tell me what the stack looks like and what worries you. If the audit is the wrong tool, I'll say so on the call.