SESSION

Transcribed Audios Automatic Checklist Based on Generative AI

Accept Cookies to Play Video

OVERVIEW

EXPERIENCEIn Person
TYPELightning Talk
TRACKGenerative AI
INDUSTRYFinancial Services
TECHNOLOGIESAI/Machine Learning, GenAI/LLMs, MLFlow
SKILL LEVELIntermediate
DURATION20 min
DOWNLOAD SESSION SLIDES

Daily, we receive 3000 audios from the Customer Relationship Center (CRC) registering Brasilprev consultants’ interactions with clients. Manual evaluation protocols in the CRC quality area face challenges with random audio checks. Our project aimed to create an automatic checklist using generative AI, eliminating the need for manual listening to entire audios. To achieve this, we established Azure Databricks pipelines for daily audio collection, transcription using Azure cognitive services, and integration with internal customer databases. MLflow Gateway and Vector Search are used to filter relevant sentences before summarizing and validating them with Azure OpenAI's GPT-4. The retrieval of the exact minute and second a customer reports the reason to cancel the plan, for example, reduces the evaluator's need to listen to the entire audio. Transcribed audios summarized by Generative AI enable a personalized service, benefiting both CRC quality area and customer experience.

SESSION SPEAKERS

Ana Paula Oliveira Bertholdo

/Data Scientist
Brasilprev