OMV drives eco-innovation in the energy sector with a RAG chatbot on Databricks

Overview

Vienna-based energy company OMV built a RAG chatbot on Databricks Agent Bricks Custom Agents to help staff navigate over 100,000 pages of product certifications and EU regulatory data, cutting document/web search time by up to 20% and reducing vector search costs by up to 25x.

The challenge

OMV's business wanted faster access to knowledge and a smoother way to synthesize huge amounts of text and tabular data related to EU rules and regulations, but its existing data infrastructure was inadequate for the required processing, and initial generative AI experiments fell short given a poor efficiency-to-cost ratio and high costs from overusing vector search units.

The solution

OMV developed a retrieval-augmented generation (RAG) chatbot using Databricks Agent Bricks Custom Agents, powered primarily by GPT-3.5 or GPT-4, with a batch ingestion pipeline using BAAI's BGE M3 model for multilingual embeddings to index 100,000 pages of regulatory and internal documents, Databricks Model Serving endpoints for deployment, and Unity Catalog to manage access control so only authorized users could reach specific documents.

Reported business value

Databricks tools helped OMV cut document and web search times by 20% and reduce the time to incrementally add more documents, while OMV experienced a 20–25x reduction in cost for its vector search tools, giving decision-makers faster, more accurate access to critical information.