Lessons learned from deploying a machine learning model

Click for: original source

More than half a million calls go through the Talkdesk systems every day. For our clients, especially larger contact centers, it can be difficult to understand and monitor everything happening due to the amount of structured and unstructured data and the high pace at which they originate. Our main goal is to make their job easier by providing insightful dashboards about calls and the overall call center performance. By Henrique Carlos.

We started by building a prototype Machine Learning (ML) model based on Latent Dirichlet Allocation (LDA) algorithm to analyze sets of voice call transcriptions and capture which topics are being talked about. Building the model was a rewarding but challenging experience. Despite that, having the model sitting on our laptops was bringing no value to the organization. We had to take it to production so that our clients could benefit from it.

The article covers following:

  • Topic extraction pipeline
  • Integrations should be the priority
  • The focus is on delivering

Deploying the machine learning model in production brought several unexpected challenges and the team was very proud of their achievement. Focusing on early end-to-end integration allowed them to iterate fast, and by using small and valuable increments we were able to overcome all issues. Good read!

[Read More]

Tags big-data data-science