AM-BITS

Customer: Mobile phone operator

Industry: Telecommunications

Scale: 1000+ employees

Challenge: Build a cost-efficient system with massively parallel processing architecture (MPP) on the Apache Hadoop HDP/HDF platform.

Solution: We built an MPP system for statistical data collection and analytics.
We utilised the following tools:

• Storage based on the HDP platform – Hadoop Data Platform (open-source software under the Apache license)

• Real-time data collection based on the HDF (NiFi) platform – Hadoop Data Float (open-source software under the Apache license)

• A data virtualization layer – TIBCO Data Virtualization was also added to the architecture to address the issues of multi-channel data integration and holistic representation of data stored in varied sources.

Result: Multi-channel integration and holistic data presentation with Tibco Data Virtualization tools.
Linear scalability of data storage capacities based on the Dell/EMC Isilone object-oriented storage technology, cutting down the response time to 1-5 seconds.

Customer: Beverage producer

Industry: F&B

Scale: ≥1000 employees

Challenge: Improve the quality of product presentation in retail chains. Automate reporting.

Solution: A proposed all-out solution ensured:

• Online control of merchandise facing, completeness of in-stock product range and equipment technical condition.

• Automated service notification in case of a discrepancy in actual product presentation and/or the location of the refrigerator.

• A cloud-based solution was chosen for greater data analytics and visualization.

• Development of a hardware-software system that goes with sensors placed in the refrigeration equipment for ongoing key parameters monitoring.

Result: Automated data collection from sensors on refrigeration equipment.
A cloud solution for online monitoring, analysis of critical parameters and service departments notification.
Capability to exchange data with user applications.

R&D PROJECT

Challenge: Create a system to upload user videos for further analysis. The system should be capable of analysing the video automatically. The system must provide for 10 concurrent user sessions and be linearly scalable to increase the number of simultaneous users up to 500.

Solution: To solve the problem, we used state-of-the-art machine learning models. The models are based on convolutional neural networks, which allows us to speed up the learning and inference tasks by using GPU. Thus, if there are sufficiently powerful GPUs in the cluster, it is possible to achieve near real-time prediction performance (the number of frames per second processed by the model approaches the video FPS value). The module allows to mark up video (photo) datasets in parallel, which is later used for model training.

Result: We developed a multi-node linearly scalable platform for automatic video analysis using GPUs, which made it possible to significantly increase the performance per cluster node. A user interface was developed for maintaining a video library, viewing and analyzing it, as well as allowing you to upload videos and select ML models to be used for analysis.

Customer: Energy company

Industry: Energetics

Scale: ≥ 1000 employees

Challenge:

• Build a predictive model.

• Automate the prediction process.

• Decrease financial losses due to forecast errors.

• Provide the client with a detailed report on model accuracy and potential ways to increase it.

 Solution: The customer provided us with the data they used to make the forecast, namely hourly data on electricity generation and daily temperature data.

To improve the accuracy of the predictive model, we enriched it with regional weather readings and data from various sources, including open ones. This allowed us to recognize and utilize hidden data patterns.

Based on the data analysis, we built models utilizing several different architectures:

• decision trees,

• neural networks,

• metric-based algorithms.

Result: We discovered the features that influence the forecast result the most, built a prototype predictive model, analyzed viability and means to implement the ML model to solve the problem, as well as the ways to improve it, and provided the client with the respective detailed report.

AM-BITS team achieved 84% model accuracy (according to data from March to October 2021).

Customer: Call center Retail services

Industry: Retail services

Scale: 1000+ employees

Challenge: The verification of call centre customers based on voice data as a part of multi-factor authentication for security improvement.

Solution: We used deep neural architecture for comparing voice prints from a database, with high verification accuracy.

Result: We built and trained an efficient ML model using a deep neural architecture to compare incoming voice data with voice samples from the previously formed database.
The security of the authentication system has been enhanced; the time for processing and comparing voice samples has been reduced; and the estimated accuracy of verification has been improved.
Low EER (Exact Error Rate) was achieved.

Customer: Energy Company

Industry: Energy

Scale: 10 000+

Challenge: Obtain a short-term energy consumption forecast for planning purchases on the Ukrainian Energy Exchange.

Solution: While analyzing Machine Learning algorithms and methods for short-term forecasting of events, we opted for using Recurrent neural networks (RNN). RNN made it possible to process sequential spatial chains or a series of events in time.
To conduct the research and set up the test RNNs, we leveraged the open-access hourly data stats on electricity consumption in New York City alongside temperature fluctuations during this period. As a result of our analysis, we managed to get an accurate forecast for two days’ consumption rates upfront by using a 3-month range of historical data.

Result: Utilization of Recurrent neural networks (RNN) to build an effective ML-model. As a result, customers can get short-term forecasts of energy consumption based on the analysis of historical data, with a high accuracy.