LiveMLP: ML Platform for Assisting Contact Center Agents in Real-Time

Wednesday, June 14, 2023 - 1:55 pm2:50 pm

Aashraya Sachdeva, Staff Engineer-ML, Observe.AI

Abstract: 

Contact centers are essential for customer support, but managing high call volumes can lead to agent stress and high attrition rates. Traditional methods of improving performance include supervisor oversight and post-call systems that use ML to analyze recordings for behavioral and communication issues. However, these approaches have limitations in knowledge retention and product/service knowledge. A real-time system is needed to guide agents during calls, but this presents engineering challenges in throughput vs latency and fault tolerance compared to post-call systems. Real-time ML systems also differ in terms of batch vs non-batched inference and context. The talk presents a real-time ML platform that scales horizontally and ensures low latency, discusses approaches to make the system robust and stable, and demonstrates its efficacy through implementation and load testing with up to 10,000 concurrent calls. Real-world evidence shows that such a system positively impacts business metrics.

Aashraya Sachdeva, Observe.AI

Aashraya Sachdeva is a technology enthusiast who is passionate about creating accessible AI products. A Machine Learning expert with a focus on platform engineering, he has years of experience handling ML projects across data assimilation, modeling, deployment, and scaling. A graduate of IISc, Bengaluru, he is currently working as Staff Engineer, Machine Learning at Observe.AI. With his extensive experience in machine learning and platform engineering, he believes in converting research into practical products that are easy to use.

BibTeX
@conference {288265,
author = {Aashraya Sachdeva},
title = {{LiveMLP}: {ML} Platform for Assisting Contact Center Agents in {Real-Time}},
year = {2023},
address = {Singapore},
publisher = {USENIX Association},
month = jun
}

Presentation Video