
-
Client
The Сlient is a renowned jewelry house in New York City. Unique selling propositions and product quality have made the Сustomer a leader in the jewelry market and brought him worldwide fame. The customer turned to the Jelvix team to develop a massive project to transform his business and reduce the labor input of a human resource in an established workflow.
-
Business Challenge
The Customer assigned the Jelvix development team the task of creating a marketplace application based on Artificial Intelligence algorithms at a user level without heavy machinery. The Customer’s main goal was to revolutionize the jewelry industry.
-
Solution
The Jelvix development team decided to go the most reliable way for drawing the set goals into life and use a separate neural network for each task. Moreover, separate neural networks were used for the top view shots and for the side view shots. As a result, all the necessary neural networks were trained to detect objects in the frame, define the contours of the detected objects, and recognize the shapes and colors of gems. Thus, we developed a model of an application that can correctly and quickly respond to any combination of user requests.
- Location
- USA
- INDUSTRY
- Retail
- SERVICES
- Android development, Back-end development, UI/UX design, QA automation
- TECHNOLOGIES
- Mask R-CNN, Tensorflow, Keras and OpenCV for machine learning and computer vision tasks, Kotlin, Android SDK, Support Library, RxJava, Retrofit 2, OkHttp, WebSocket, Dagger 2, Moxy, Glide, Gson, Crashlytics, MPAndroidChart, Min API version: 21 (Android 5.0), Django channels for receiving and sending data between Django app and Android app; PostgresSQL, Django framework and Django REST framework, docker and docker-compose for development and deploy, Celery for asynchronous tasks with Redis as a broker, AWS s3 bucket as images storage.
Product Overview
Client’s goals
The customer turned to the Jelvix team to develop a marketplace mobile app for buying, selling, and evaluating jewelry and gemstones. The main essence of the application development was to enable any user to evaluate, download, buy, sell, or assess the authenticity of gems and jewelry. The functional part of the platform was to replace the work of the jewelers and save users’ funds.

Implementation
The first stage in the project's development was to determine the number of functional tasks of the platform. Having identified this aspect, our development team proceeded to the next stage, precisely developing a separate neural network for each task. This stage was the longest and the most difficult.
The next step was to define and create classifiers, establish relationships between them, and check the accuracy of the results after their use.
At the next and final stage, the project was given to our QA automation team to make sure that the platform and all its components were working accurately.
Value Delivered
Having brought the Customer's project to life, the Jelvix project team brought his business to a new level of interaction between customers, jewelers, and sellers. As a result of the platform's implementation, employees' labor intensity dropped significantly, while the number of successful deals increased.

Project Results
The result of our team's work is the whole creation of an MVP version of the mobile application for Android. The app can take and analyze a photo of a gemstone and give the results in the form of an assessment of its dimensions, size, color, and imperfections. The main functions of the application are performed quickly and accurately.