ClusterCraft: Automated Clustering for Parts Catalogs

Home Page

Inspiration

Krish’s father—a consultant for Airgas—introduced us to a major challenge: Airgas’s parts catalog contains thousands of items that aren’t grouped together, even when they’re essentially the same item in different sizes. Inspired by Amazon’s proprietary tagging algorithm, we set out to create a solution that brings order to this chaos.

What it does

Our platform leverages ML-driven clustering to automatically group similar parts, turning disorganized catalogs into coherent, searchable collections that streamline inventory management. Additionally, we've integrated an intelligent chatbot that provides instant, detailed answers to any product inquiries, making it easier than ever to navigate and understand your catalog.

How we built it

Data Collection: We scraped over 300 items from the Airgas website, gathering essential details such as item titles, descriptions, prices, Airgas part numbers, manufacturer part numbers, and image URLs. Preprocessing: We meticulously cleaned and organized the raw data into relevant categories, ensuring consistency and accuracy for further analysis. Clustering: Utilizing the K-means algorithm, we determined the optimal number of clusters using the elbow method, effectively grouping similar items together. Frontend Visualization: To bring our data to life, we developed an interactive frontend with Next.js, allowing users to visually explore and navigate through the organized clusters.

Challenges we ran into

We experimented with various techniques—including DBSCAN and fuzzy matching (fuzzywuzzy)—and discovered that effective preprocessing is crucial for model performance.

Accomplishments that we're proud of

We successfully categorized all items into their correct groups, validating the accuracy and robustness of our approach.

What we learned

This project deepened our expertise in web scraping, data preprocessing, and model fine-tuning. It also taught us the value of iterative testing and patience when working with complex datasets.

What's next for AI-Powered Parts Clustering for Smarter Inventory Management

We aim to extend our model to other item catalogs, addressing a critical need across part catalog companies like Airgas, Snap-On, and beyond, while exploring advanced algorithms to further enhance performance.