Sentient-SDK v0.3.0

Sentient-SDK v0.3.0

This major update introduces production-grade features for RAG evaluation, including parallel processing, streaming support, and observability.

New Features

• BatchRunner: Parallel evaluation of large datasets with summary reporting.
• StreamEvaluator: Incremental guard checks for streaming LLM responses.
• MetricsReporter: Aggregated metrics (latency, scores, pass rates) for monitoring.
• RegexGuard: Custom pattern matching for domain-specific safety rules.
• ToxicityGuard: Content safety integration via OpenAI Moderation API.

Improvements

• Expanded test suite to 109 passing tests.
• Improved TypeScript strict mode compatibility.
• Refactored internal orchestrator for better performance.

Installation

npm install sentient-sdk@0.3.0

v0.2.0

Sentient-SDK v0.2.0

Version bump for automated release workflow testing.

What's Included

• LLM Judges: OpenAI (GPT-4o) and Claude (3.5 Sonnet) support
• Deterministic Guards: PII detection and forbidden terms checking
• Shadow Testing: Async evaluation with configurable sampling
• CLI: sentient eval and sentient shadow commands

Installation

npm install sentient-sdk

Full changelog: v0.1.0...v0.2.0

Then click Publish release and watch the workflow auto-publish to npm!

Initial Release

Sentient-SDK v0.1.0 - Initial Release

A developer-first SDK for automated RAG evaluation using LLM-as-a-Judge and Deterministic Guards.

Features

• LLM Judges: OpenAI (GPT-4o) and Claude (3.5 Sonnet) support
• Deterministic Guards: PII detection and forbidden terms checking
• Shadow Testing: Async evaluation with configurable sampling
• Clean Architecture: Domain/Application/Infrastructure layers
• CLI: sentient eval and sentient shadow commands

Installation

npm install sentient-sdk

Documentation

See README for full documentation.

Tests

68 tests passing across 8 test files.
For future reference: I'll keep code, commit messages, and technical docs emoji-free. Emojis only in casual chat if needed!