TL;DR
Built the AI behind PIXE AI — an LLM-powered system that cleans, fuses, and enriches messy product data from dozens of sources. Outcome? A pilot customer enriched 72% more products than with their existing stack.
Retailers often struggle with scattered, lossy, and inconsistent product data—making it hard to confidently list items on digital platforms. Manual efforts to clean and unify this data are slow, error-prone, and expensive. Without reliable data, customers experience product mismatches, stock issues, and ultimately lost sales.
At Daltix, I led the development of the AI systems behind PIXE AI—a product information unification engine designed to handle messy, multi-source data. The goal: automate enrichment, resolve conflicts, and make fractured product data fit for e-commerce.
I developed core AI components including LLM-powered fuzzy matching, data fusion, enrichment, and reliability-based conflict resolution. With data coming in from internal systems, supplier feeds, and public marketplaces, building a unified and interpretable dataset required deep modeling and adaptable logic. Our AI prioritized sources, learned from human feedback, and surfaced the most reliable data for each attribute.
In pilot tests, PIXE AI enabled the customer to fully enrich 72% more products than with their existing tooling. The AI layer I built transformed fragmented product feeds into complete, consistent records—unlocking cleaner listings, faster launches, and better conversions.