Caterpillar2024 – 2026
Finance data platform
A GenAI text-to-SQL tool that lets thousands of finance employees explore ERP data in Snowflake conversationally, the pipelines that feed it, and a separate training platform for Caterpillar's move to OneStream.
- Role
- Finance Data Platform Engineer
- Stack
- SnowflakePythonSQLAWS GlueSnapLogicNext.jsPostgreSQL
Text-to-SQL
Finance runs on ERP data, but most people who need an answer can't write the SQL to get it. So they file a request and wait, or export to a spreadsheet and hope they got the filters right.
I led development and rollout of a general-purpose text-to-SQL tool, now used by several thousand finance employees a month. The model was never the hard part. What mattered was making the data legible to it: optimizing the schema, curating what the agent can see, fine-tuning the agent, and testing real finance workflows end to end until the answers held up.
- finance employees using it monthly
- 1000s
- legacy source systems unified
- 30+
Underneath it, I built batch and near-real-time pipelines with Python, SQL, AWS Glue, SnapLogic, and Snowflake, unifying more than 30 SAP, Oracle, IBM DB2, Teradata, and other legacy sources.
OneStream training platform
When Caterpillar adopted OneStream as a new finance system, thousands of people needed to learn it. I built a separate Next.js training platform for that rollout, with database-backed progress tracking and prerequisite logic so people learn things in the right order. More than 3,000 employees have used it.