How Pulumi built an AI-powered analytics assistant that lets any employee query data using natural language.

The Pulumi blog provides developers with resources, tutorials, and best practices for infrastructure as code (IaC) and cloud-native development using Pulumi's open-source tools. Developers can learn about deploying and managing infrastructure using familiar programming languages such as JavaScript, TypeScript, Python, and Go. Additionally, the blog covers topics such as cloud architecture patterns, multi-cloud deployments, and serverless computing, enabling developers to build and deploy modern cloud applications with ease.

Pulumi

Pulumi built an internal AI analytics assistant that lets employees query their data warehouse using natural language. The key insight: a semantic layer (using Cube) that defines business metrics and data relationships is more critical than the LLM itself. Without it, AI-generated SQL produces confidently wrong results through incorrect joins, missing filters, and wrong aggregations. The semantic layer constrains the AI to use pre-defined measures and dimensions, making entire categories of errors structurally impossible. The system is deployed on AWS with Pulumi IaC and accessible via web UI, Slack bot, and Model Context Protocol server. Since launch, it handled 1,700+ queries from 83 employees, shifting the data team from answering routine questions to platform work.

How We Built Platybot: An AI-Powered Analytics Assistant

The problem with throwing AI at your Data Warehouse