Our team will reach out within 24 hours to gather your project requirements, clarify your business objectives, and outline the next steps in our collaboration.
Client
Glia is a B2B interaction platform specializing in facilitating communication between financial companies like banks, credit unions, and insurance companies and their customers.
Challenge
Glia generates millions of client call minutes monthly. Analyzing these calls in real time posed a significant business challenge. The company needed an optimal, technically efficient, cost-effective solution to improve customer support and care.
Solution
DataArt’s team deployed the optimized open-source Whisper model to Amazon Sagemaker inference endpoints for near real-time transcriptions. To conduct sentiment analysis, they used Claude-instant on AWS Bedrock, categorizing call transcripts segments as negative, neutral, or positive. The solution was integrated into Glia’s platform as a widget, providing operators with a real-time sentiment dashboard. The transcripts and sentiment labels are securely stored in the platform for further analysis.
AWS Lambda (SAM) was utilized to handle serverless compute tasks, ensuring efficient processing. Amazon S3 provided scalable storage for raw audio files and processed transcripts. Amazon EventBridge Scheduler was employed to automate and manage the workflow, ensuring timely processing of call data. Additionally, Amazon RDS for PostgreSQL was used for structured data storage and management.
The most time-consuming part of the process was selecting the optimal speech-to-text model for real-time call processing. Through extensive research and testing, the team found Whisper model performed best on English datasets for call transcription. Also, the fastest inference engine was chosen.
Technology
AWS Bedrock
Amazon RDS for PostgreSQL
Amazon DynamoDB
Amazon EventBridge Scheduler
Amazon S3
Amazon Lambda (SAM)
Outcomes
- Achieved approx. 12-sec speech-to-text latency, and 13-16 sec sentiment analysis latency.
- Transformed the solution into an applet and fully integrated it into Glia’s platform.
- Equipped operators with access to all the available data on the platform and real-time call sentiment analysis, enabling more efficient customer support.
