How I Built a Self-Correcting Data Engine for my AI Vault
From the Butcher to the Critic: I built a three-pass audit system to clean 180 legacy blueprints. Here is the exact logic I used.
In Part 1, I showed you how I built the “House”—the glass-morphism dock and the architectural UI of the Prompt-Led Vault. I focused on the aesthetic because, in product, your interface is the handshake. It’s the first sign of technical credibility.
But a pretty interface is just a hollow shell if the data inside is hallucinated.
When I exported my 180+ Substack posts, I realized that asking an AI to “summarize this into a database” resulted in generic garbage. The AI was lazy. It stripped the “Engineer” out of my writing and left me with corporate fluff. It treated my technical blueprints like generic blog posts.
I had to architect a sort of a Multi-Pass Logic Pipeline. This is how I forced an AI to think like a Senior PM and why I spent my Friday afternoon fixing a routing disaster that nearly killed the login system for my paid subscribers.
The “Concentration Window” Problem
Most people treat LLMs like they have infinite focus. They don’t. If you shove a 2,000-word technical post into a prompt and ask for a 4-part schema mapping, the model will prioritize the middle of the text and “forget” the edge cases at the end. To me, this the Concentration Window.
When the AI loses focus, it starts to “normalize” your data. It replaces my specific technical variables with phrases like “optimizing the workflow.” That is a death sentence for a Logic Engine like Prompt-led Product Vault.
To fix this, I stopped using a single prompt. I built a Three-Pass Audit System to process the 180 legacy files:
Pass 1 (The Butcher 🔪): Strip all HTML and marketing fluff to leave only the raw technical argument.
Pass 2 (The Architect 👷♂️): Map that argument to my strict 4-part schema (Problem, Logic, Prompt, Strategy).
Pass 3 (The Critic 👩⚖️): A final “Zero-Latency” check to ensure no technical variables were lost in translation.
I Break AI Tools So You Don’t Have To
If you’re building a product and want a Senior PM to find the exact point where your users give up, I’m taking on new clients. I’ll check your architecture and test your logic before your users do. You can book a Product Audit here.