Telecommunications|Product Engineering|Increase Efficiency

LTIMindtreeCall summarization

Manual fine-tuning held back deployment. Automated optimization cut inference time 63% and agent routine work by over 50%.

Jan 19, 2026|17 days ago

Key results

Response Time
~7 secs
vs 16.75 seconds
Saved Per Call
5-10 mins
vs manual summary extraction
Agent Time Reduction
>50%

The company

LTIMindtree logo

LTIMindtree

ltimindtree.com

Technology consulting and digital solutions provider for global enterprises.

IndustryTelecommunications
LocationMumbai, MH, India
Employees10K-50K
Founded1996

Result highlights

  • Chatbot response time cut from ~17s to ~7s
  • 5-10 mins saved per call for telecom clients
  • >50% reduction in agent time for telecom clients

The story

A global technology consulting firm providing AI-powered contact center solutions to the telecommunications industry.

Customizing AI models for individual clients required manual fine-tuning that created operational bottlenecks and slowed deployment. Conventional tuning methods like grid search were too inefficient to optimize for both summarization accuracy and the low latency required for real-time interactions.

The firm integrated the SigOpt Intelligent Experimentation Platform to automate hyperparameter tuning for its BART-based transformer models. The system uses Bayesian optimization to balance ROUGE scores against inference time, generating a Pareto Frontier that helps engineers select the ideal configuration for each client. This workflow runs on Intel Xeon processors, replacing manual search methods with an automated pipeline.

Scope & timeline

  • 50% fewer iterations for model optimization

Quotes

Explore similar

Find AI opportunities for your
business context

Understand what's working with 2,383 recent AI case studies across industries. We structure things so you can find high-impact strategies for your exact context.

Graphic placeholder

Industries covered