• Designed multi-stage AI-assisted classification workflows transforming unstructured regulatory data into a structured, multi-level product taxonomy.
• Built human-in-the-loop validation systems that route low-confidence classifications to expert review, significantly reducing manual workload.
• Designed semantic search and recommendation experiences improving recall discovery speed and relevance.
• Created confidence-based UX patterns and guardrails to balance automation with trust in a consumer-facing regulatory product.
• Partnered closely with ML engineers, product, and compliance teams to align model behavior with real-world regulatory constraints.
• Optimized system performance, reducing classification time from 16–18 hours to 3–4 hours through tiered model strategy.
Impact: Delivered a largely autonomous recall monitoring experience while preserving accuracy, auditability, and user trust.