Post by Jennifer Alfonso

The domain was wine. The methodology was engineering. 🍷⚙️ To demonstrate applied AI fluency for my return to tech, I spent a year running a structured experiment using 10+ LLMs to solve a high-variance problem: mapping subjective human sensory data to objective production fingerprints. I didn't just ask for recommendations; I built a 300+ SKU dataset to benchmark model consistency and identify repeatable failure modes. 3 Reasons why AI Program Managers should read this: Hallucination Triggers: I documented a typo that propagated across 3 model iterations, leading to confident, non-existent recommendations. Context Stripping: Without domain framing, a stateless model classified wine names as "California real estate principals". Architecture Mapping: I identified which models excel at deep inference vs. fast pattern matching for complex workflows. This is a case study in Root Cause Analysis applied to a new class of tools. This analysis is also available as a downloadable PDF in the Featured section of my profile. Full methodology paper and failure analysis below. 👇 #TechnicalProgramManagement #AI #LLMs #DataIntegrity #Engineering