Hawaiian Electric builds a RAG chatbot on Databricks to answer regulatory compliance queries in seconds

Overview

Hawaiian Electric Company (HECO) worked with Databricks Professional Services to build a retrieval augmented generation proof of concept using Databricks AI Search and a RAG model-serving endpoint, letting its regulatory team query extensive regulatory documentation through a conversational chatbot instead of manually reading full texts. The implementation took two weeks to go from zero to a working RAG system and cut the time to answer a single regulatory query from about five minutes to five seconds.

The challenge

HECO's regulatory team had to manually sift through a vast trove of regulatory documentation, searching for relevant documents and reading through entire texts to answer information requests, a repetitive and time-consuming process that hindered the team's ability to respond quickly and at scale.

The solution

HECO partnered with Databricks Professional Services to build a RAG model proof of concept, setting up a model-serving endpoint for querying regulatory documents, a Databricks Notebook interface for iterative query refinement, and Databricks AI Search with automatically generated Spark Declarative Pipelines to ingest preprocessed data into a vector database, with Unity Catalog securing non-public documents.

Reported business value

The RAG proof of concept was implemented in two weeks and reduced the time to answer a single regulatory query from about five minutes to five seconds, improving both efficiency and the accuracy and reliability of the information retrieved, with HECO planning to scale the chatbot to other departments.