About the Client

eJohri is India’s first omnichannel jewelry marketplace, offering a seamless blend of online and offline shopping. The platform allows customers to purchase jewelry online or locate nearby retail stores for in-person purchases. Featuring jewelers from metropolitan, urban, and suburban areas, eJohri showcases a wide variety of jewelry, including gold, silver, kundan, platinum, zirconia, and solitaire.

Country:

India

Industry:

Retail

Download PDF

Business Situation and Requirements

The jewelry retail industry is undergoing rapid digital transformation, with marketplaces striving to meet growing consumer expectations for fast, seamless, and visually engaging shopping experiences. For eJohri, India’s first omnichannel jewelry marketplace, one critical challenge was the time-intensive process of editing product images. On average, editing a single jewelry image manually took 3–4 hours, slowing down onboarding for new jewelers and limiting the platform’s ability to scale efficiently while maintaining high-quality visuals for gold, silver, platinum, and zirconia products.

To overcome this, eJohri sought a technology partner to develop an AI-based image editing solution that could automate and accelerate the workflow. Unthinkable was entrusted with designing and implementing a machine learning-driven platform capable of bulk editing images while preserving product quality, color, and texture. The solution integrated seamlessly with eJohri’s existing infrastructure, reduced image processing time from hours to seconds, and supported additional features such as single product–multiple seller management and category-wise commission setup, helping the marketplace scale operations, improve efficiency, and enhance the overall customer experience.

Their key requirements included:

Select Suitable AI Technology – Evaluate and implement the most effective AI/ML techniques to handle bulk image processing efficiently.
Seamless Platform Integration – Ensure the AI-based solution integrates smoothly with the platform, supporting scalable onboarding and ongoing operations.
Augment Image Dataset – Expand the existing dataset to train the ML model for higher accuracy and better handling of diverse jewelry images.
Maintain Image Quality – Preserve the original clarity, color, and details of the jeweler’s images to reflect products authentically.
Automate Image Editing – Replace the manual 2–3 hour editing process per image with an AI-driven system to accelerate jeweler onboarding.
Reduce Onboarding Time – Ensure that jewelers can quickly upload and list their products, minimizing delays in going live on the platform.

The Solution

To modernize and scale the e-commerce experience for the client, we developed an AI-driven solution that automates image processing while strengthening marketplace operations. By combining advanced machine learning, transfer learning techniques, and seamless platform integration, the solution transforms manual editing into a fast, accurate, and scalable workflow. Alongside intelligent image enhancement, we implemented robust marketplace features such as multi-seller product listings and flexible commission management, creating a high-performance, automated ecosystem that enhances visual consistency, accelerates seller onboarding, and optimizes revenue operations.

AI-Powered Image Editing

The platform uses machine learning to fully automate the manual 2–3 hour image editing process per jewelry item. It detects object boundaries, removes backgrounds, standardizes image dimensions, and optimizes lighting and color, ensuring consistent presentation across all products. This reduces human dependency, accelerates jeweler onboarding, and allows rapid scaling as more sellers join the platform.

Optimized ML Model Selection

Various computer vision techniques, such as edge detection, image thresholding, and K-means clustering, were evaluated but proved inconsistent across different jewelry types. A supervised machine learning model was implemented instead, providing uniform accuracy across a wide variety of designs, including intricate patterns, gemstones, and metal finishes. This ensures precise image processing regardless of jewelry complexity.

Data Augmentation via Transfer Learning

To overcome limited datasets, transfer learning was applied using pre-trained convolutional neural networks (CNNs) as the base. Techniques like image rotation, scaling, flipping, and color jittering expanded the dataset, enabling the model to learn variations in jewelry images and achieve high accuracy across diverse product types.

High Accuracy and Reliability

The trained ML model achieved a 95% accuracy rate in editing jewelry images while preserving key product attributes like color, shine, texture, and intricate design details. This guarantees that the edited images remain faithful to the jeweler’s portfolio, maintaining trust and visual consistency on the platform.

Seamless Integration with Platform

The AI-based solution integrates directly with the eJohri platform using RESTful APIs and batch processing pipelines. It supports bulk image uploads, real-time processing, and scalable onboarding, ensuring that the entire workflow from image submission to live product listing is streamlined and operationally efficient.

Single Product, Multiple Sellers

Multiple sellers can list the same jewelry product under a single product ID. This feature enables buyers to compare pricing, view aggregated reviews, and select the best offer, while the system maintains accurate inventory tracking and updates for each seller. It simplifies product management and improves the buyer’s decision-making experience.

Category-wise Commission Setup

Admins can define commission rates for different product categories, allowing flexible revenue management across the platform. The system automatically calculates seller payouts based on category-level rates, supports overrides for individual products, and integrates seamlessly with the platform’s accounting and payout workflows, ensuring transparency and efficiency.

The Impact

The AI-powered image editing and optimized ML model enabled eJohri to increase image editing efficiency by 98%, reducing manual editing time from 3–4 hours to just 2–3 seconds per image. Data augmentation, transfer learning, and high-accuracy processing ensured product images retained original quality, color, and texture across the platform.

Seamless platform integration, single product–multiple seller support, and category-wise commission setup streamlined onboarding, improved inventory management, and simplified seller payouts, accelerating platform scalability and enhancing overall operational efficiency.