Client background
The client manages a highly sophisticated satellite network and connectivity infrastructure, ensuring seamless coverage for customers worldwide.
Published client profile: Industry: Telecom. Partnership period: January 2021 - present.
Business challenge
The client sought to streamline the processing of multilingual customer support logs to improve issue identification, assess support quality, and accelerate problem resolution. By leveraging generative AI and Large Language Models (LLM), the company aimed to automate various processes, eliminate manual tasks, and reduce troubleshooting time and costs, ultimately enhancing service quality and boosting customer satisfaction.
Implementation
The delivery team helped the client develop a solution that utilizes two fine-tuned LLMs to generate summaries from multilingual customer support chat logs, displayed through an intuitive dashboard.
The first model provides a comprehensive summary of the customer conversation, highlighting the main issue and potential resolution. The second model filters out irrelevant requests to ensure only relevant data appears in the dashboard. Both models support multiple languages (e.g., Spanish, Portuguese) and provide English translations.
Additionally, the models automatically classify chat logs into predefined topics and a set hierarchy. Paired with the dashboard, the solution allows the client to efficiently search, filter, and analyze customer feedback for specific services, significantly reducing manual effort and enabling the team to focus on higher-priority business tasks.
The new genAI solution comprises the following architecture components:
Apache Airflow serves as a job orchestration tool.
Amazon EMR runs queries against the source table and stores the results in Amazon S3.
Amazon SNS is automatically triggered when a file lands in S3, then invokes a Lambda function.
The Lambda function triggers batch transform jobs, powered by Amazon SageMaker.
Amazon SageMaker provides model hosting, deployment, and inference.
The output data is stored in Amazon S3.
The Tableau report is generated based on the output data.
A Docker image is created for the Sagemaker transform job processing and hosted on Amazon ECR.
All Sagemaker logs are stored in AWS CloudWatch.
Value delivered
The AI solution enables the client to identify and mitigate issues faster. By leveraging generative AI to summarize customer support requests automatically, the client has accelerated troubleshooting by up to 40%.
The solution also offers advanced data analytics capabilities, allowing the client to process vast amounts of data within minutes and make effective, timely decisions.
Moreover, the AI solution automatically translates support logs from Spanish and Portuguese into English, making it easier for the client to provide top service to customers worldwide.
Expertise delivered: Generative AI development services Technologies: Apache Airflow, Amazon EMR, Amazon S3, Amazon SNS, Amazon SageMaker, Tableau, Amazon ECR, AWS CloudWatch, Docker 40% faster troubleshooting Faster processing of large amounts of data Automatic inquiry translation into English
