JayDevs

Web application for virtual product presentation. Allows potential buyers to familiarize themselves with product appearance and consumer properties using virtual reality.

Business Case

Quarantine measures and restrictions on the movement of people during the COVID pandemic resulted in a significant drop in sales for shop owners selling general goods. Buying something significant potential customers still wanted to thoroughly study product characteristics and appearance, consult a shop assistant which was impossible without visiting the store. 

Showroom

A virtual showroom for products was created as a solution to address the need enabling customers to walk between the stands with goods, study their design in detail, get comprehensive information about the characteristics of the goods, and consultant a virtual assistant on demand.

Solution & Functionality

• The application is a client-server solution where the client part provides a 3D showroom space implemented using the Unity3d technology. The choice of technology was determined by the business requirement to provide ultimate photorealism and maximal consumer immersion in VR.
• First-person navigation and on-demand presentation of additional stand elements with product details were implemented in VR space.
• A virtual assistant (chatbot) was built into the showroom along with a shopping cart enabling adding items without leaving the showroom.  The IBM Watson service was utilized as an assistant implementation technology to create chatbots equivalent to Siri and Google Assistant having specific knowledge about the product though.
• The backend was used as a repository of 3D models and all the content required for the operation of the virtual showroom.
• The web application was integrated into the online merchandise trading platform and launched when viewing the product card.

#JavaScript #HTML5 #Unity3D #IBM Watson 

The module that is based on machine learning algorithms predicts sales of future periods, growth/decline trends for certain groups of goods.

Business Case

The store assortment has a certain value for categories and subcategories of goods. It was required to make a prediction of sales for each subcategory over a period of 3 months.

Initial Information

1. Daily sales for the certain period (date, product, product group, product balance at the end of the day, number of sales).
2. Content of Receipts (daily for the period): receipt, date, product, type of discount, percent of discount, card number, quantity, amount, discount amount, retail price category, type of loyalty card.
3. Description of goods and product categories product, product name, brand product group, subcategory, supplier.

Solution & Functionality

The dataset with the time series of sales by product subcategories was prepared based on the initial dataset. Various approaches to forecasting time series have been tested for modeling purposes: ARIMA, ARCH, recurrent neural networks, etc. At each step, the symmetric mean absolute percentage error (SMAPE) based on model predictions and actual sales over the last 3 months was calculated. As a result, the decision to use an additive regression model was made.

Seasonality of demand for goods was identified for each subgroup:

• Annualy
• Monthly
• Weekly

In addition to that:

• The trend per each product group was determined
• Weekly sales forecasts for a tree-month range were made
• Prediction accuracy for 75% of sales comprised over 85% (SMAPE)

#Recurrent neural networks #ARIMA #ARCH #Time Series Model

Layout control automation. Merchandisers take pictures of shelves with products. The system recognizes photos, compares to a reference and automatically forms the list of goods available on stock, checks assortment correctness and proportion of occupied shelves compared to competitors.

Business Case

The goal of the project was to reduce human participation while improving the quality of work performed in retail outlets.

Monitoring of goods layout on the shelves allows evaluating assortment correctness depending on the object type and proportion of occupied shelves compared to competitors. Further, it can be leveraged for order automation within integration with a warehouse.

Solution & Functionality

• Recognition of goods on photos with the help of CNN neural networks.
• CNN - is a deep learning algorithm able to take an incoming image, assign values ​​(learning weights and biases) to various image aspects and objects as well as identify one from another.
• Image recognition models are complemented by a predictive analytics module.

Recognition accuracy: 97%

Recognition speed: ~2 min

Number of learning images: ~300000 (300 per product)

#Python #Tensorflow #JavaScript #HTML5

A machine learning algorithms based module that predicts customer behavior when they receive advertising content (SMS / Telegram / Viber).

The task of the module is to generate a list of buyers interested in a particular mailout in order to reduce the number of "unnecessary" messages and increase conversion.

Business Case

Using machine learning and customer data the objective was to create a product that would:

• save money and assist in budget management;
• assist in making a marketing strategy;
• increase conversion;
• allow avoiding unnecessary interruptions of a customer.

Maximum Optimization

Utilizing the available customer data (purchase history and behavior during sales) the objective was to ultimately optimize sending promotional offers.

Having identified favorable moments for sending messages, the module helps to save on delivery of "unnecessary" messages + increase customer loyalty.

Solution & Functionality

The module is a cloud solution using the CatBoost library. Neural network training is based on the user database. The following parameters were used to create the algorithm: 

• Purchase frequency, seasonality (annual, monthly, weekly, daily), date of last purchase;
• Average purchase volume, favorite products;
• Activities in promotions: in which participated, the discount amount, ignored promotions;
• Loyalty card based buyer’s profile: gender, age, personal data;
• Synthetic data (Generation of datasets based on existing business processes. Applying various distortion methods to existing data).

The module revealed certain data patterns based on which further decisions were made. Behavior prediction accuracy comprised 80-82%.

#eCommerce #Healthcare #CatBoost #Python #iOS #Android 

Face recognition system that uses biometry to match facial features on photo or video within several seconds. With the purpose of identification or authentication, the recognition technology compares obtained data with the existing database to find the match.

Business Case

For security reasons access to the building and separate rooms is granted only to certain people. Biometry-based recognition allows organizing a system of quick access that does not require additional actions from visitors. The system also enables visitors’ location control.

Solution & Functionality

The biometric system is a client-server application and works in four steps:

• Photography
• Saving
• Matching
• Acceptance

When uploading a photo (video), general information about a person (name, surname, passport data) is entered into the database. Then the face is analyzed and a unique identifier is generated. To generate the identifier the MobileNetv3 architecture based neural network trained on 200,000 photos is used. During each verification, the generated identifier is compared with the stored identifier. Then the system either confirms or denies the access right.

#Python #Tensorflow #JavaScript #HTML5 #Security

The system determines different types of protective equipment and if they are used correctly.

Business Case

For production facilities, there is a need to control the proper use of protective equipment to minimize accident occurrences. Cameras record the working process. Videos from them are viewed by the neural network in real-time mode to observe the compliance of wearing personal protective equipment.

If a violation is detected an operator/shift supervisor is immediately informed to address the issue promptly.

Solution & Functionality

The system is aimed at recognizing protective equipment - headgear (helmet) and protective clothing (gloves, boots) as well as the correctness of use (working clothes are fully buttoned, sleeves are not rolled up, etc.)

Cameras monitor personnel during work and flag violations.

The system is a client-server solution consisting of:

• A media server processing video from cameras;
• A neural network based on MobileNetv3 architecture, trained on 250,000 videos, is used to detect violations in images;
• Notification modules;
• Report generation.

#Savety&Health #Construction #Python #Tensorflow #JavaScript #HTML5

Embedded web widget that automatically creates a 3D model of a room based on its 2D plan. Enables users to transform the obtained 3D plan in terms of colors/textures and allows them to navigate through the final virtual space.

Business case

A study of conversions on real estate sales/lending platforms-aggregators demonstrated that having an object 3D plan in the announcement increases the chance of closing a deal. 

The majority of sellers and landlords in the secondary housing market don't have 3D plans and have to either buy them from designers, which considerably increases the cost of ads placement or miss the opportunity of making their announcements more attractive.

2D to 3D

The widget recognizes a plan issued by the real estate registration bureau or a plan on the facility certificate and recreates the 3D object scene considering scale. All the elements on the plan are taken into consideration: doors, windows, ventilation shafts, etc.

The Server Side:

• Room plan is recognized and transformed into an object structure applying neural network based on MobileNetv3 architecture taught on 200K of marked up plans of houses and apartments; 
• If needed the plan image quality can be improved utilizing OpenCV library algorithms.

The Client Side:

• Presents a 3D image of floor/walls considering doorways, windows, and other elements of the plan. 3D space is realized based on the ThreeJS framework that provides a great speed to reality ratio requiring no installation of additional browser components;
• Transforms wall and floor colors/textures alternating object 3D models in real-time mode. Scene objects are not pre-baked;
• Provides first-person navigation using ThreeJS integrated functionality.

The widget automatically supports CSS styles of the site it is embedded in if realized applying recognized HTML standards.

#RealEstate #Rental #Web #Mobile #3D #Responsive

Embedded widget for online stores enabling the dynamic demonstration of textiles with a quality close to photorealistic. Allows visualizing how potential textile might look in motion with different lighting.

Sample Widget

Product presentation matters a lot when selling textiles and accompanying goods. Fabric on a plain surface looks differently than pleated and flowy. Online sales of textiles make the choice even more challenging when a buyer has no opportunity to see and sense product samples live. 

A widget presenting fabric in motion was developed to enable the online presentation of textiles. A user can change lighting parameters, the number of pleats and motion range to dynamically visualize textiles.

The widget allows demonstrating textiles with exquisite texture at photorealistic scale. To realize this the system of tile (unique textile elements) positioning on curtain surface was designed.

Solution & Functionality

• The widget functionality is based on an adjustable 3D scene realized based on the ThreeJS framework. The choice of the framework was determined by the business’s need for high working speed without installing additional components and plugins.
• A curtain 3D model, being a set of points each of which is an atomic physical entity having its parameters and space orientation, is the central object of the scene.
• At the same time, the color of each entity is set based on the textile texture pattern and color considering rapport that allows users to percept curtains as a single object.
• Realization of the widget assumes automatic site style application to the widget directly employing standard CSS.
• The widget is absolutely self-contained: it gets only texture from the site while all the controls of lighting, scale, etc. are realized utilizing UI provided by the widget.

#eCommerce #Web #Visualization #3D 

The application for automated reading, storage and dimension analyses of a heart rate monitor that being designed specifically for older people. AI performs an early detection of possible heart problems before appearing obvious signs through the analysis of heart rhythm abnormalities.

Heart rate device

Predicts health problems business case The measurement heart rate device for elderly people does not have a large screen because of the size, weight, an increase in battery life requirements. The existing screen is suitable for displaying only current measurements, which is the user experience significant limitation.

Predicts health problems

For more convenient devise use and successfully competing with other manufacturers, our client requested the development of its own IT products ecosystem. It is a set of applications that synchronize with a heart rate monitor. Also, it collects and stores measurement results analysis and predicts possible health problems of the wearable device owner.

The application system:

• Android application version that synchronizes with the device and collects indications in real-time. The application also implements “User Account” functionality with the ability to bind to the current user in the system or to the user social network profile. Server synchronization is carried out in a secure connection mode.
• Web application version that provides full access to the measurement results, convenient reports and analytics visualization, as well as limited access to other personal data through the personal account owner, invite.

Technical implementation:

• Data storage and API for access to it with differentiation by roles;
• GPU cluster, on which neural networks are being launched and carry out predictive analytics based on the measurement results;
• All applications communicate with each other via secure channels in order to avoid the loss of personal users’ data.

 #Heathcare #AI #IoT #cloud #Android #ReactJS

The app is a fun and easy way to indulge, save and shop. A mobile-commerce platform for smart shoppers of today, bringing you a wide selection of carefully chosen deals, catering to popular genres like restaurants and pubs, spas, salons, tattoos studios, shopping deals, fun activities and lots more, all at the lowest possible prices.

Discounts aggregator of different brands

In today’s era where consumers are common people that are always looking for options where they can save as much as possible on shopping goods and services. Discounts aggregator of different brands, allowing to receive notifications on discounts and offers from the brands on the move and when you pass nearby them. The app helps to reach the desired shop by using mobile application navigator. Under slick UI the interactive business logic of the app attracts spoiled customers.

Technical implementation:

• Authorization and registration module;
• Displaying lists of coupons;
• Displaying coupons on Google map;
• Viewing companies;
• Push notifications;
• Integration with social networks and instant messengers: Google, Facebook, Twitter, WhatsApp;
• Coupons activation by scanning a barcode;
• Coupons activation by detecting beacons, registered by these coupons.

#eCommerce #iOS #Android #IoT

SDK is a set of components that allows mob and web developers to expand their product's functionality with photo/video transformation tools. SDK allows to remove the existing object or place a new one in real-time.

SDK allows eCommerce product developers to improve household goods' visualization. It's possible to build customer Mob Apps, Web Apps, tablet Apps, AR glasses that will improve/ simplify the process of furniture and interior objects “fitting”, increasing electronic commercial conversion.

SDK can work both in a client-server app mode and fully autonomously if the device itself supports this feature.

SDK includes the mechanism of image quality adaptation accordingly to the device capacity for optimal balance quality/speed.

The project ecosystem and its technical implementation description:

• GPU cluster for running neural networks and data transformation (Python, OpenCV, Keras, Tensorflow);
• Components for IOS (AR Kit, Swift, C); Components for Android (AR Core, Java, Kotlin, C ++);
• Unity3d plugin for cross-platform development (C #, native iOS and Android libraries).

#eCommerce #AI #machine vision #IOS #Android #React

The solution is designed for marketers working with the audience in Ad networks (Adwords, Facebook, etc ). 

The configuration tool helps to manage the audience and deliver specific data to the target customers. 

Solution is used to automate day-to-day activities of marketers in Ads domain.

The tool provides the possibility to create Audience delivery configuration. The workflow engine compiles Audience information from different sources and delivers it to different Ad networks. Apache Spark is used to compile and deliver Audience data. It allows to process significant data sets (70M - 200M records) within reasonable time.

Workflow returns statistical information about each reaction/activity to the Configuration tool.

Functionality:

• Management of Audience delivery configurations;
• Advanced Audiences search functionality;
• Collecting Audience delivery statistics;
• Separation of workspaces between users;
• User management & security.

#BigData #AdWords #BI #Hadoop #Go #Python #Spark

ProSurvey automates survey participation, enabling you to submit data spending less time and effort.

With ProSurvey’s intuitive online experience,

you can upload your company data, check suggested job matches, review guided alerts with immediate data verification, and keep up-to-date with instant visibility of work-in-process. And participation gets easier year after year with using of machine-learning algorithms.

Developed functionality:

• Opportunity to share information about yourself and your company through the Profile Setup wizard;
• Uploading the company’s employees data file with a click or drag-and-drop;
• Job matching: automatic and manual. If the application doesn’t find an exact match, it will offer a list of suggested matches;
• Automatic reviewing and data cleaning - prompting you to validate specific information as necessary;
• Progress of your submissions;
• Advanced filter options to see your progress by country, company, or overall;
• SSL encryption for all API requests.

#platform #BigData #RabbitMQ #NodeJS #Angular #microservises

The solution designed for medical centers provides an easy and convenient way for patients to book appointments with dentists. Mobile applications for dentists provide simple and convenient management of orders and information about patients

Dental healthcare app 

 Mobile applications for dentists provide simple and convenient management of orders and information about patients portfolio JD Mobile applications for dentists provide simple and convenient management of orders and information about patients. The application allows dentists to view and manage assigned, current and previous appointments with access to post-care notes and recommended treatments.

Solution designed for medical centers 

This is the app that’s developed to save money, time and potentially a lot of pain too. The app is a whole new way of looking at dentistry, one that is almost entirely from a patient’s point of view. Designed to deliver convenience, value, transparent pricing and great dental care.

Functionality implemented within Android & iOS apps:

• Ability to create / update users' profiles;
• Appointment booking for family members;
• Various payment methods integrated ( Stripe, PayPal , Apple /Google pay , Medicare, Afterpay)
• After care recommendations;
• Push notifications for reminding patients about their appointment & when dentist is running late;
• Feedback about the dentist/practice;
• Integration with Google Maps API, Apple Maps API;
• SSL encryption for all API requests.

#Healthcare #Platform #Peer-to-peer #iOS #Android

Cloud solution for searching, publishing and collaborating on the graphical models, regardless of the type of the modeling tool.

The web platform unifies documents and models to facilitate access, management and tracking of versions and dependencies between models. 

With this plugin, the user can easily add charts to their document and keep it up to date.

Platform benefits:

• -Easy web access to all models;
• Tool and content type independent traceability;
• Powerful analytics;
• Tight integration with chosen tools.

Solution

Developers have implemented a unique set of components based on the latest version of the modern JavaScript framework Angular, which allows to significantly improve the visualization of complex data models, processes and interaction with users in different browsers.

The main task of the client side is models’ mapping, their dependencies, the history of changes. Among interesting challenges there is the analysis of possible dependencies while changing the initial model or model element. All potentially affected models (in case of changes in the primary model) needed to be found and displayed on the client in the form of a table.

#BPM #Java #Angular #SaaS #Docker

The Module predicts customer behaviour using machine learning algorithms. It forms the list of recommended activities in CRM for the most efficient interaction with every potential customer.

Business Case

MIA Module is used to search for the right target group for the product/service.

It’s also used for effective product/ service sales strategy resource intensive process. Sometimes the market or product/service conjuncture changes require to experiment with the target group and sales approach, even if the processes are running well-functioning processes. This is a time and money consuming work/process, which is more than relevant in the case of organizing sales from scratch.

This solution is an extension of Zoho-CRM. Using patterns that were founded in historical data, the Module predicts the probability of conversion in the chain “a cold lead- a hot lead- a client”. The Module provides tools for sales pipeline analysis and current client’s portrait. (how close it’s to the target one). This instrument allows prompter and deeper tracking and changes evaluation. It reduces costs and improves customer base quality.

The module is a cloud solution consisting of the following components:

• A cascade of neural networks and statistical algorithms for analyzing customer data has been developed;
• Server for performing computational operations and launching neural networks;
• Extension for Zoho-CRM with the ability to install via marketplace;
• Itegration with Zoho-CRM API is Implemented;
• SSL encryption is enabled for API requests;
• The reporting functionality is Implemented in Zoho-CRM;

#eCommerce #CRM #Machine Learning #Java #Python #Keras

The Odds On application is a feature-rich prototype designed to test how users will play and interact in the competitive setting of sports betting.

This easy-to-understand application gives access to the realistic sport betting competition that uses real odds from real odds markets in real time and in various sports. This way the application provides the user with the opportunity to understand real betting strategies and get pleasure in a competitive game.

Developed functionality:

• Home page with video guide;
• Leader board displaying the list of all participants and their winnings. Users can also check bets of other players;
• Club chat with group and 1-1 chats;
• Push-notifications;
• Competition page where user can ‘Make a pick’ from different available sports;
• Sport details page with events' list for a selected sports category.

#Sport #mobile #Swift #iOS