A Revolution in Customer Service with Advanced AI and Supercomputing

In Bucharest, Romania, the startup RepsMate developed a SaaS solution that integrates AI & Data Analytics technologies to understand customer needs and behavior. With this information, call center agents can better tailor their communication to meet customer needs. 

RepsMate aims to provide customer service representatives at organizations with automated, customized guidance, offering a real-time, always-available virtual coach. With deeper insights into customer needs and behavior, managers can make more informed decisions. NCC Romania and NCC Netherlands collaborated to support RepsMate in optimizing their service.   

Technical/Scientific Challenge of RepsMate 

The technical solution of RepsMate is based on various functionalities:  

• Automatic speech recognition 
• Spoken emotion recognition
• Behavior segmentation (1) 
• Contextual real-time guidance (2) 

All these elements extensively utilize current machine learning technologies. These technologies enable the processing of raw information and the extraction of detailed insights from the emotions and behavior of the speaker. This information is key to improving service quality. 

The main challenge is to ensure the optimal implementation of all elements. Additionally, it is important to test the coherence of the application’s building blocks so that it can run on large-scale computing machines. 

Collaboration 

The collaboration between NCC Romania, NCC Netherlands, and RepsMate focused on analyzing the audio processing workflow performed by RepsMate, as well as adapting the code written for the software for the purpose of fast-processing and parallel computation on the GPU. 

This initial screening and evaluation of the RepsMate solution were conducted by consultants from NCC Romania. The screening revealed that the application had a clear need for high-performance computing. 

Therefore, RepsMate developers were given access to the Dutch national supercomputer Snellius to run some tests on Nvidia GPUs. They also received assistance from experts at NCC Netherlands for general performance evaluation and verification. 

The result? The analysis proved the scalability of RepsMate and its efficient use on large-scale systems with Nvidia GPUs. 

More tests will need to be conducted to see if the application also works on AMD GPU systems. This will ensure the workflow is scalable across both major GPU suppliers. To perform these tests, we are assisting RepsMate in applying for development access to the most suitable pre-exascale system in Europe (funded by EuroHPC-JU). 

The Impact for RepsMate 

The need for supercomputing was justified by the high degree of parallelization required for the current AI solution, which uses deep learning. This deep learning employs large convolutional neural networks (very smart computer networks) with specific classifiers. Additionally, a large number of audio files must be processed, making it ideal for high-performance computing. 

 This assessment showed that the AI application of RepsMate is well-implemented and can optimally use HPC systems (high-performance computing). This will undoubtedly help accelerate the time-to-solution when working with specific customer internal knowledge bases. This can lead to more satisfied clients, and in the end, better results for the company or organization.  

In particular, the in-depth analysis of the AI application helps to understand the best way to run RepsMate on various systems with minimal costs, both financially and in terms of sustainable computer system usage. 

What has RepsMate learned from using HPC?   

• A concrete description of how the application runs on a supercomputer
• Verification of the optimal processing for the application 
• Efficient use of computing resources on medium-scale test runs 
• Guidelines for using large-scale test runs for future use of the largest EuroHPC pre-exascale systems in Europe 

About NCC Romania 

NCC Romania is hosted by the National Institute for Research & Development in Informatics – ICI Bucharest, the leading institute in research, development, and innovation in information and communication technology (ICT) in Romania. 

About NCC Netherlands 

NCC Netherlands is hosted by SURF, the ICT cooperative that supports Dutch educational and research institutions in developing and/or acquiring the best possible digital services to promote knowledge sharing through continuous innovation. SURF hosts the national supercomputing and storage facilities of the Netherlands. 

1. Behaviour segmentation  is  a technique for the segmentation of user behaviour or customer behaviour. 

1. Real–time guidance on call centers is immediate feedback and suggestions to individuals as they perform a task or activity.