Spaces:
Sleeping
Sleeping
File size: 11,863 Bytes
135f0d6 159faf0 135f0d6 159faf0 135f0d6 159faf0 135f0d6 159faf0 135f0d6 a52e676 135f0d6 a52e676 135f0d6 159faf0 135f0d6 159faf0 135f0d6 159faf0 135f0d6 159faf0 135f0d6 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 |
"""
Enhanced RAG Pipeline with Guardrails Integration
This module extends the existing RAG pipeline with comprehensive
guardrails for response quality and safety validation.
"""
import logging
import time
from dataclasses import dataclass
from typing import Any, Dict, List, Optional
from ..guardrails import GuardrailsResult, GuardrailsSystem
from .rag_pipeline import RAGConfig, RAGPipeline, RAGResponse
logger = logging.getLogger(__name__)
@dataclass
class EnhancedRAGResponse(RAGResponse):
"""Enhanced RAG response with guardrails metadata."""
guardrails_approved: bool = True
guardrails_confidence: float = 1.0
safety_passed: bool = True
quality_score: float = 1.0
guardrails_warnings: Optional[List[str]] = None
guardrails_fallbacks: Optional[List[str]] = None
def __post_init__(self):
if self.guardrails_warnings is None:
self.guardrails_warnings = []
if self.guardrails_fallbacks is None:
self.guardrails_fallbacks = []
class EnhancedRAGPipeline:
"""
Enhanced RAG pipeline with integrated guardrails system.
Extends the base RAG pipeline with:
- Comprehensive response validation
- Content safety filtering
- Quality scoring and metrics
- Source attribution and citations
- Error handling and fallbacks
"""
def __init__(
self,
base_pipeline: RAGPipeline,
guardrails_config: Optional[Dict[str, Any]] = None,
):
"""
Initialize enhanced RAG pipeline.
Args:
base_pipeline: Base RAG pipeline instance
guardrails_config: Configuration for guardrails system
"""
self.base_pipeline = base_pipeline
self.guardrails = GuardrailsSystem(guardrails_config)
logger.info("EnhancedRAGPipeline initialized with guardrails")
def generate_answer(self, question: str) -> EnhancedRAGResponse:
"""
Generate answer with comprehensive guardrails validation.
Args:
question: User's question about corporate policies
Returns:
EnhancedRAGResponse with validation and safety checks
"""
start_time = time.time()
try:
# Step 1: Generate initial response using base pipeline
base_response = self.base_pipeline.generate_answer(question)
if not base_response.success:
return self._create_enhanced_response_from_base(base_response)
# Step 2: Apply comprehensive guardrails validation
guardrails_result = self.guardrails.validate_response(
response=base_response.answer,
query=question,
sources=base_response.sources,
context=None, # Could be enhanced with additional context
)
# Step 3: Create enhanced response based on guardrails result
if guardrails_result.is_approved:
# Use enhanced response with improved citations
enhanced_answer = guardrails_result.enhanced_response
# Update confidence based on guardrails assessment
enhanced_confidence = (base_response.confidence + guardrails_result.confidence_score) / 2
return EnhancedRAGResponse(
answer=enhanced_answer,
sources=base_response.sources,
confidence=enhanced_confidence,
processing_time=time.time() - start_time,
llm_provider=base_response.llm_provider,
llm_model=base_response.llm_model,
context_length=base_response.context_length,
search_results_count=base_response.search_results_count,
success=True,
error_message=None,
# Guardrails metadata
guardrails_approved=True,
guardrails_confidence=guardrails_result.confidence_score,
safety_passed=guardrails_result.safety_result.is_safe,
quality_score=guardrails_result.quality_score.overall_score,
guardrails_warnings=guardrails_result.warnings,
guardrails_fallbacks=guardrails_result.fallbacks_applied,
)
else:
# Response was rejected by guardrails
rejection_reason = self._format_rejection_reason(guardrails_result)
return EnhancedRAGResponse(
answer=rejection_reason,
sources=[],
confidence=0.0,
processing_time=time.time() - start_time,
llm_provider=base_response.llm_provider,
llm_model=base_response.llm_model,
context_length=0,
search_results_count=0,
success=False,
error_message="Response rejected by guardrails",
# Guardrails metadata
guardrails_approved=False,
guardrails_confidence=guardrails_result.confidence_score,
safety_passed=guardrails_result.safety_result.is_safe,
quality_score=guardrails_result.quality_score.overall_score,
guardrails_warnings=guardrails_result.warnings + [f"Rejected: {rejection_reason}"],
guardrails_fallbacks=guardrails_result.fallbacks_applied,
)
except Exception as e:
logger.error(f"Enhanced RAG pipeline error: {e}")
# Fallback to base pipeline response if available
try:
base_response = self.base_pipeline.generate_answer(question)
if base_response.success:
# Create enhanced response with error warning
enhanced = self._create_enhanced_response_from_base(base_response)
enhanced.error_message = f"Guardrails validation failed: {str(e)}"
if enhanced.guardrails_warnings is not None:
enhanced.guardrails_warnings.append("Guardrails validation failed")
return enhanced
except Exception:
pass
# Final fallback
return EnhancedRAGResponse(
answer=(
"I apologize, but I encountered an error processing your question. "
"Please try again or contact support if the issue persists."
),
sources=[],
confidence=0.0,
processing_time=time.time() - start_time,
llm_provider="error",
llm_model="error",
context_length=0,
search_results_count=0,
success=False,
error_message=f"Enhanced pipeline error: {str(e)}",
guardrails_approved=False,
guardrails_confidence=0.0,
safety_passed=False,
quality_score=0.0,
guardrails_warnings=[f"Pipeline error: {str(e)}"],
)
def _create_enhanced_response_from_base(self, base_response: RAGResponse) -> EnhancedRAGResponse:
"""Create enhanced response from base response."""
return EnhancedRAGResponse(
answer=base_response.answer,
sources=base_response.sources,
confidence=base_response.confidence,
processing_time=base_response.processing_time,
llm_provider=base_response.llm_provider,
llm_model=base_response.llm_model,
context_length=base_response.context_length,
search_results_count=base_response.search_results_count,
success=base_response.success,
error_message=base_response.error_message,
# Default guardrails values (bypassed)
guardrails_approved=True,
guardrails_confidence=0.5,
safety_passed=True,
quality_score=0.5,
guardrails_warnings=["Guardrails bypassed due to base pipeline issue"],
guardrails_fallbacks=["base_pipeline_fallback"],
)
def _format_rejection_reason(self, guardrails_result: GuardrailsResult) -> str:
"""Format user-friendly rejection reason."""
if not guardrails_result.safety_result.is_safe:
return (
"I cannot provide this response due to safety concerns. "
"Please rephrase your question or contact HR for assistance."
)
if guardrails_result.quality_score.overall_score < 0.5:
low_quality_msg = (
"I couldn't generate a sufficiently detailed response to your "
"question. Please try rephrasing your question or contact HR "
"for more specific guidance."
)
return low_quality_msg
if not guardrails_result.citations:
return (
"I couldn't find adequate source documentation to support a response. "
"Please contact HR or check our policy documentation directly."
)
return (
"I couldn't provide a complete response to your question. "
"Please contact HR for assistance or try rephrasing your question."
)
def get_health_status(self) -> Dict[str, Any]:
"""Get health status of enhanced pipeline."""
base_health = {
"base_pipeline": "healthy", # Assume healthy for now
"llm_service": "healthy",
"search_service": "healthy",
}
guardrails_health = self.guardrails.get_system_health()
overall_status = "healthy" if guardrails_health["status"] == "healthy" else "degraded"
return {
"status": overall_status,
"base_pipeline": base_health,
"guardrails": guardrails_health,
}
@property
def config(self) -> RAGConfig:
"""Access base pipeline configuration."""
return self.base_pipeline.config
def validate_response_only(self, response: str, query: str, sources: List[Dict[str, Any]]) -> Dict[str, Any]:
"""
Validate a response using only guardrails (without generating).
Useful for testing and external validation.
"""
guardrails_result = self.guardrails.validate_response(response=response, query=query, sources=sources)
return {
"approved": guardrails_result.is_approved,
"confidence": guardrails_result.confidence_score,
"safety_result": {
"is_safe": guardrails_result.safety_result.is_safe,
"risk_level": guardrails_result.safety_result.risk_level,
"issues": guardrails_result.safety_result.issues_found,
},
"quality_score": {
"overall": guardrails_result.quality_score.overall_score,
"relevance": guardrails_result.quality_score.relevance_score,
"completeness": guardrails_result.quality_score.completeness_score,
"coherence": guardrails_result.quality_score.coherence_score,
"source_fidelity": (guardrails_result.quality_score.source_fidelity_score),
},
"citations": [
{
"document": citation.document,
"confidence": citation.confidence,
"excerpt": citation.excerpt,
}
for citation in guardrails_result.citations
],
"recommendations": guardrails_result.recommendations,
"warnings": guardrails_result.warnings,
"processing_time": guardrails_result.processing_time,
}
|