Databricks applies GenAI to improve higher education student advising

Databricks applies GenAI to transcribe calls at scale using models like OpenAI Whisper for higher fidelity. An LLM-as-a-judge approach scores advisor performance against institutional rubrics, enabling targeted reviews and replacing random sampling.

The solution enriches call data with sentiment, topics, and intent analysis. Insights are surfaced through an Agent Bricks Knowledge Assistant for unstructured reasoning over transcripts and a Genie Space for structured trend queries.

All stages of the solution, from data ingestion and transcription to AI analysis and discovery, run on a single governed platform. Unity Catalog ensures data security, and AI Functions allow direct LLM calls from SQL.

Educational institutions face challenges in scaling the review of call center quality for adviser services, involving manual transcription and analysis of conversations, which is costly and difficult to perform at scale.

The solution involves deploying Whisper on Databricks Model Serving, using AI Functions for enrichment, and orchestrating with LangGraph. It can be deployed using Databricks Marketplace for Whisper models and Foundation Model APIs for LLMs, with compute attached to single-user clusters or serverless environments.