Generative AI for catalog normalization and search optimization
From unstructured descriptions to an intelligent, accurate catalog optimized for search.
The Challenge
A catalog fed with free-text descriptions, errors and inconsistent naming made product classification difficult, reducing the accuracy and the performance of the search engine.
The Solution
A hybrid system of rules, fuzzy matching and generative AI that automatically normalizes and classifies descriptions, turning them into structured data to optimize the catalog and its searches.
Key capabilities
- Automatic normalization of free-text descriptions.
- Classification by type, family and subfamily.
- A combination of rules, fuzzy matching and generative AI.
- Detection of new parts for validation.
- Synchronization with OpenSearch.
- Continuous auditing of the accuracy of the system.
Technology Stack
Impact
- A significant improvement in search performance, with average latency down from 8,227 ms to 2,162 ms in testing.
- A new search engine 3.8 times faster than the previous approach.
- Overall accuracy of around 93.58% on the evaluation set, combining rules and AI.
- A cost of approximately $0.43 per 1,000 parts processed in batch.
- A better user experience, with faster, more accurate and more consistent filters.
- A foundation ready for more autonomous and scalable inventory cataloguing.