LLM-powered data classification for data entities at scale

Summary

Grab needed to classify sensitive data at the table and column level across petabytes of database tables and streaming schemas. Manual schema-tiering campaigns had resulted in half of all schemas receiving overly strict Tier 1 access controls, while an initial automated service using regex patterns and third-party machine learning produced high false-positive rates and lacked customizability. To address this, the Caspian data engineering and governance teams enhanced their internal orchestration service, Gemini, by integrating GPT-3.5 via Azure OpenAI. Gemini aggregates classification requests into mini-batches, handles API rate limits, and uses prompt engineering—including few-shot examples, curated tag libraries, and explicit JSON DTO schemas—to reliably tag columns for data owner verification.

Takeaways

1. 1

   Schema-level sensitivity tagging overly restricted access because a single Tier 1 table caused an entire schema of hundreds of tables to inherit the strictest access controls.

2. 2

   Gemini relies on message queues to aggregate scanning requests into mini-batches and applies workflow-level rate limiting to adhere to Azure OpenAI limits of 240K tokens per minute.

3. 3

   Prompt engineering techniques such as providing JSON DTO schemas, few-shot examples, and explicit default fallback tags (<None>) ensure the LLM returns structured, machine-parseable classifications.