You are viewing a single comment's thread from:

RE: LeoThread 2025-10-18 14-48

in LeoFinance2 months ago

Part 6/10:

  • Implements prompt engineering strategies, including:

  • Extraction of contributing objects.

  • Identification of attributes like columns and data sources.

  • Explicit handling of various query types, e.g., lineage, impact analysis.

This layered approach ensures that the AI understands context and provides precise, trustworthy responses.

Model Selection and Scaling

Initial experiments utilized GPT-4 with a 32,000-token context window but faced limitations with extremely large stored procedures exceeding token limits. Transition to GPT-4 Turbo (with 128,000-token capacity) drastically improved performance, requiring fewer iterations and achieving 85-93% accuracy in lineage extraction.

Performance Benchmarks