AWS: Halliburton AI assistant for seismics cuts workflow creation time by over 95 percent

AWS published a case study on Halliburton, the oilfield services giant, which built a generative AI assistant for its Seismic Engine together with the AWS team. The goal was to translate natural language into executable seismic workflows and lower the barrier for engineers without deep geophysical expertise.

How is the assistant technically composed?

The system runs on Amazon Bedrock with two Claude models — Sonnet 3.5 V2 and Haiku 3.5 — alongside Amazon Nova Lite for intent routing. Amazon Bedrock Knowledge Bases provides a RAG layer over 82 specialized Seismic Engine tools, while conversation history is stored in DynamoDB. The application runs on AWS App Runner, with vector search handled by OpenSearch Serverless. The user describes a workflow in natural language and the assistant assembles it from the available tools.

What figures does Halliburton report?

Workflow generation success rates range from 84 percent (Claude Haiku 3.5 on simple cases) to 97 percent (Claude Sonnet 3.5 V2 on medium-complexity tasks). Creation time dropped from 2–20 minutes of manual work to 5.9 to 16.6 seconds of generation, which the company describes as “more than 95 percent reduction in time.” Halliburton’s executive in the announcement emphasizes that the solution “not only improves efficiency and accuracy, but also makes advanced geophysical tools accessible to a broader range of users” — a key advantage in a sector facing a serious shortage of senior talent.