Trendyol Launches Three Open-Source AI Models

Trendyol, one of Turkey’s leading e-commerce platforms, has launched three innovative open-source artificial intelligence (AI) models aimed at revolutionizing the online shopping experience. These models, focused on image processing, were developed using Trendyol’s extensive e-commerce data and infrastructure and are now publicly available through the Hugging Face platform (Swipeline, 2025).

With millions of daily transactions and product interactions, Trendyol has emphasized the importance of leveraging AI to improve product discovery, catalog management, and the overall quality of its visual content. By releasing these models as open-source, the company aims to encourage wider adoption and collaboration within the AI and e-commerce communities, enabling developers worldwide to enhance their platforms with cutting-edge technology.

DinoV2 Image Similarity Model

The DinoV2 Image Similarity model has been fine-tuned specifically for e-commerce use cases. It enables more accurate and faster visual search by helping users find products that look similar to the ones they are interested in. This model supports Trendyol’s recommendation engines and visual search tools by improving product matching, which directly contributes to better customer engagement and satisfaction.

Visual search is rapidly gaining traction in online retail, as it allows customers to search using images rather than text, which is often more intuitive and efficient. DinoV2’s ability to distinguish fine visual details between products enhances user experience and can increase conversion rates by simplifying the discovery process (Swipeline, 2025).

E-Commerce Product Image Encoder

Built on the ConvNeXt architecture, the E-Commerce Product Image Encoder model helps detect duplicate or visually similar products within Trendyol’s vast catalog. Managing product duplicates is a critical challenge for large e-commerce platforms as it affects search quality and inventory management.

By accurately encoding product images into rich feature vectors, this model aids in clustering similar items, improving the platform’s ability to organize product catalogs, filter redundant listings, and present more relevant search results. This enhancement not only benefits the backend operations but also improves the shopper’s journey by reducing confusion caused by duplicate entries (Swipeline, 2025).

Background Removal Model

The third model, an optimized version of the IS-Net architecture, focuses on automatically removing backgrounds from product photos, particularly for fashion and portrait images. Clean and professional product images are essential for e-commerce success, as they influence buying decisions and overall brand perception.

This background removal model simplifies and accelerates the photo editing process by allowing bulk removal of distracting or inconsistent backgrounds. The result is a more uniform and visually appealing product presentation, which can enhance the quality of listings and ultimately drive higher sales (Swipeline, 2025).

Insights from Dr. Tolga Ahmet Kalaycı

Dr. Tolga Ahmet Kalaycı, Data Science Director at Trendyol Group, elaborated on the company’s AI development philosophy: “At Trendyol, we are continuously addressing challenges related to catalog quality, content moderation, and the semantic representation of our products. To effectively solve these issues at scale, we go beyond off-the-shelf AI solutions, developing and customizing models tailored to our specific needs. These models are deployed in live production environments, processing millions of transactions every day” (Swipeline, 2025).

His statement highlights Trendyol’s commitment to innovation and quality in both product data management and customer experience, driving the company’s competitive edge in the fast-growing e-commerce market.

The Significance of Open-Source AI

By making these AI models open-source, Trendyol is not only advancing its own technological capabilities but also contributing to the global AI ecosystem. This transparency and willingness to share cutting-edge tools promote collaboration among researchers, developers, and other companies in the retail sector.

Open-source models encourage faster innovation cycles, allowing external developers to adapt, improve, or build upon Trendyol’s technology for diverse use cases. This can accelerate the adoption of AI-driven solutions across various industries, fostering a more dynamic and inclusive tech environment (Swipeline, 2025).

Future Outlook

Looking ahead, Trendyol plans to continue enhancing its AI portfolio and expanding the practical applications of these technologies within its platform. The company aims to leverage AI to streamline operations, enhance personalization, and offer customers a seamless shopping experience.

As the e-commerce landscape becomes increasingly competitive, investments in AI-driven automation and improved data quality are essential to meet evolving consumer expectations. Trendyol’s proactive approach positions it well to remain a leader in the region while contributing to the broader AI and retail innovation communities.