2M+
Product Listings Indexed
Case study · Marketplace
AI-Powered Product Search for Commerce Marketplace
Intelligent product search, image-to-text catalog processing, and a seller rewards ecosystem for a leading multi-vendor commerce marketplace.
Drupal
PHP
Elasticsearch
SolrOCRFigmaPhotoShopProject details
- Client
- Multi-vendor commerce marketplace
- Industry
- E-Commerce / Marketplace
- Timeline
- 20 weeks
- Platforms
- Custom build on Drupal
The challenge
What we walked into.
The marketplace had grown to hundreds of sellers listing products across dozens of categories. But the product catalog was expanding faster than the infrastructure could handle. Finding the right product meant scrolling through endless lists with basic keyword search that missed context entirely.
A significant portion of product listings were image-only -- sellers photographed labels, packaging, and product sheets rather than typing descriptions. None of this visual content was searchable. Buyers knew the products existed but couldn't find them.
Manual catalog moderation was a bottleneck that grew worse with every new listing. Product data quality varied wildly, and the platform had no way to automatically assess or score listing completeness. They also lacked incentive infrastructure to reward sellers for high-quality catalog content and sustain marketplace economics.
What we built
The engineering decisions.
Image-to-Text Product Catalog Pipeline
We built an image recognition pipeline that processes every product listing -- extracting text from photographed labels, packaging, and product spec sheets. The pipeline normalizes extracted product data, tags it with category metadata, and feeds it directly into the search index. Products that were previously invisible became instantly discoverable.
Dual Search Engine Architecture
Rather than choosing between Elasticsearch and Solr, we deployed both in a complementary configuration. Elasticsearch handles full-text search with relevance scoring across the product catalog. Solr powers the faceted shopping experience -- filtering by category, brand, price range, seller rating, and product attributes. The combined system delivers product discovery that is both relevant and precisely filterable.
Seller Rewards & Marketplace Credits
We designed and built a complete marketplace credits system within the Drupal platform. Sellers earn credits for uploading complete, high-quality product listings; credits can be applied toward featured placement and reduced commission rates. The system handles transactions, balances, transfer rules, and anti-fraud measures -- creating a self-sustaining marketplace economy.
AI Catalog Quality Scoring
Every product listing is scored automatically based on description completeness, image quality, pricing accuracy, and category relevance. High-quality listings surface higher in search results, and sellers with consistently high scores earn bonus marketplace credits. This replaced the manual moderation bottleneck and created a catalog quality flywheel.
The results
Measured outcomes.
2M+
Product listings indexed and searchable
3x
Search accuracy improvement
10K+
Marketplace credit transactions / month
- 2M+ product listings fully indexed and searchable in real time, including previously invisible image-only listings
- Search accuracy improved 3x over the previous keyword-only system with multi-faceted product filtering
- Marketplace credits system processing 10K+ transactions per month, creating a self-sustaining seller economy
- Automated catalog moderation reduced manual review workload by 80%, letting the team focus on marketplace growth
Next case study
Retail Inventory OptimizationHave a similar challenge?
We'd love to hear about it. Every conversation starts with a senior engineer who's built systems like yours.