Skip to content

Latest commit

 

History

History
31 lines (24 loc) · 1.23 KB

File metadata and controls

31 lines (24 loc) · 1.23 KB

Changelog

[1.2.0] - 2024-12-22

Added

  • LLM-as-Judge: Advanced context validation using OpenAI/Claude for semantic accuracy checking
  • Developer Mode: Strict code grounding validation for technical codebase queries
  • ContextValidator: Dedicated service for context validation with LLM Judge and word-matching fallback
  • developerMode config option to enforce strict code reference validation
  • useLLMJudge config option to enable/disable LLM-based context validation

Changed

  • Increased LLM Judge context window from 1000 to 3000 characters per source
  • Enhanced validation prompts with mode-specific rules (Developer vs User mode)
  • Improved confidence scoring to exclude disabled validation checks from weight calculation
  • Updated CLI test to include LLM Judge and context validation options

Fixed

  • Dynamic confidence weight calculation when some checks are disabled
  • Validation fallback behavior when API keys are not provided

[1.1.0] - 2024-12-19

Added

  • Initial release with core validation features
  • Query classification for skipping non-knowledge queries
  • Accuracy checking via LLM verification
  • Hallucination detection
  • Confidence scoring with breakdown
  • Support for OpenAI and Claude providers