McKinsey
Medium
Guesstimate the annual demand for automated vacuum cleaners in India
Problem 1 you raised: "The two diagrams don't connect."
Fixed. The narrative now has three clear acts:
Act 1 (Slide 7): "This is what happens when a developer calls your API." Just the 4-stage lifecycle. No safety yet. The viewer understands the journey first.
Act 2 (Slides 8-12): "Now where can things go wrong?" Walk through each stage one by one. Each slide highlights one stage, grays out the rest, and shows a red "Failure Point" badge with the specific failure. Input fails because of jailbreaks. Context fails because PII hides in RAG docs. Model fails because it is probabilistic. Output fails because harmful content gets through. Then slide 12 adds the two OUTSIDE the lifecycle (before: who is calling? after: what patterns emerged?). Payoff line: "4 lifecycle failures + 2 outside = 6 stages."
Act 3 (Slide 13): The complete 6-stage map. Now both diagrams make sense because one was derived from the other.
Problem 2 you raised: "Solution is only for one thing."
Fixed. There are now 6 solution slides (slides 14-19), one for every stage:
Stage 0: 3-tier access system
Stage 1: Two-pass input classification
Stage 2: System prompt protection + PII scanning + re-injection
Stage 3: Two-layer policy (immutable + configurable)
Stage 4: Sync/async output split
Stage 5: Aggregate detection + red teaming + retraining loop
Community Answers (0)
No answers yet. Be the first to share your approach!