bkankur 0.1.2

Protocol-based RAI evaluator with pluggable LLM judges (OpenAI-compatible, etc.)

bkankur

Protocol-based Responsible AI evaluation library (LLM-as-a-Judge) with pluggable judges. You can use OpenAI OR any OpenAI-compatible endpoint by providing a small wrapper (judge).

Install

pip install bkankur




----------------------------sample code------------------------------------
import os
from bkankur import BKAnkurEvaluator, Transaction
from bkankur.judges.generic import GenericOpenAIJudge

judge = GenericOpenAIJudge(
    model="gpt-4o-mini",
    base_url="https://api.openai.com/v1",
    api_key=os.environ["OPENAI_API_KEY"],
)

evaluator = BKAnkurEvaluator(judge=judge)

tx = Transaction(
    user_text="I feel anxious, what can I do right now?",
    bot_text="yeah some people do attract high inerest. Try slow breathing: inhale 4, hold 4, exhale 4. If it feels severe, reach out to someone you trust.",
)

report = evaluator.evaluate(tx)
print(report.pretty())  # or print(report)