daef 0.1.0

Domain-Aware Evaluation Framework — intelligent LLM evaluation with domain-specific metrics

DAEF — Domain-Aware Evaluation Framework

Intelligent LLM evaluation that adapts to your domain. Instead of generic metrics, DAEF researches your domain, understands industry-specific requirements, and generates tailored evaluation criteria using a multi-agent pipeline powered by Google's Agent Development Kit.

Installation

pip install daef

Quick Start

import asyncio
from daef import DAEFClient, EvaluationRequest

client = DAEFClient(api_key="YOUR_GOOGLE_API_KEY")

request = EvaluationRequest(
    domain="Healthcare",
    task_description="Medical Q&A chatbot answering patient questions",
    task_type="single_call",
    focus_areas=["Security and Guardrails", "Content Generation Quality"],
    prompt="What is the recommended dosage of ibuprofen for adults?",
    llm_output="The standard adult dose of ibuprofen is 200-400mg every 4-6 hours...",
)

result = asyncio.run(client.evaluate(request))
print(f"Overall Score: {result.overall_score}/100")
for metric in result.metrics:
    print(f"  {metric.metric_name}: {metric.score}/100")

Features

• Domain-aware: Researches domain-specific standards (Healthcare, Legal, Finance, Education, and more)
• Adaptive metrics: Selects from 30+ built-in metrics and supports custom metrics
• Task-type support: RAG, Fine-tuning, Single LLM Call
• Version comparison: Compare two evaluation results to understand regressions and improvements
• Async-first: Built on asyncio for high-throughput pipelines

Configuration

Set your Google API key via environment variable or constructor:

export GOOGLE_API_KEY="your-key"

# Or pass directly
client = DAEFClient(api_key="your-key")

EvaluationRequest Fields

Field	Type	Required	Description
domain	str	Yes	Domain (e.g. "Healthcare", "Finance")
task_description	str	Yes	What the LLM task does
task_type	str	Yes	"rag", "tuning", or "single_call"
focus_areas	list[str]	No	Up to 3 priority areas
prompt	str	No	The input prompt sent to the LLM
llm_output	str	No	The LLM's response to evaluate
context_data	str	No	Retrieved context (for RAG)
mandatory_metrics	list[str]	No	Metrics that must be included
avoided_metrics	list[str]	No	Metrics to exclude
custom_metrics	list[str]	No	User-defined metric names

Version Comparison

result_v1 = asyncio.run(client.evaluate(request_v1))
result_v2 = asyncio.run(client.evaluate(request_v2))

comparison = asyncio.run(client.compare(result_v1, result_v2))
print(f"Overall change: {comparison.overall_change}")
print(f"Summary: {comparison.summary}")

License

MIT