Spaces:
Sleeping
Sleeping
Upload config.py
Browse files- src/config.py +123 -0
src/config.py
ADDED
|
@@ -0,0 +1,123 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import os
|
| 2 |
+
from dotenv import load_dotenv
|
| 3 |
+
|
| 4 |
+
# Load environment variables
|
| 5 |
+
load_dotenv()
|
| 6 |
+
|
| 7 |
+
class Config:
|
| 8 |
+
"""Configuration class for Insurance Helper RAG application"""
|
| 9 |
+
|
| 10 |
+
# API Keys
|
| 11 |
+
GEMINI_API_KEY = os.getenv("GOOGLE_API_KEY")
|
| 12 |
+
QDRANT_API_KEY = os.getenv("QDRANT_API_KEY")
|
| 13 |
+
QDRANT_URL = os.getenv("QDRANT_HOST")
|
| 14 |
+
|
| 15 |
+
# Gemini Model Configuration
|
| 16 |
+
GEMINI_MODEL = "gemini-2.5-flash"
|
| 17 |
+
GEMINI_TEMPERATURE = 0.2
|
| 18 |
+
GEMINI_MAX_OUTPUT_TOKENS = 2048
|
| 19 |
+
|
| 20 |
+
# Embedding Configuration
|
| 21 |
+
EMBEDDING_MODEL = "models/text-embedding-004"
|
| 22 |
+
EMBEDDING_DIMENSION = 768
|
| 23 |
+
|
| 24 |
+
# PDF Processing Configuration
|
| 25 |
+
CHUNK_SIZE = 1000
|
| 26 |
+
CHUNK_OVERLAP = 200
|
| 27 |
+
|
| 28 |
+
# Chunking separators optimized for insurance documents
|
| 29 |
+
SEPARATORS = [
|
| 30 |
+
"\n\n", # Paragraph breaks
|
| 31 |
+
"\n", # Line breaks
|
| 32 |
+
". ", # Sentence breaks
|
| 33 |
+
", ", # Clause breaks
|
| 34 |
+
" ", # Word breaks
|
| 35 |
+
"" # Character breaks
|
| 36 |
+
]
|
| 37 |
+
|
| 38 |
+
# Qdrant Configuration
|
| 39 |
+
COLLECTION_NAME = "insurance_documents"
|
| 40 |
+
VECTOR_SIZE = EMBEDDING_DIMENSION
|
| 41 |
+
DISTANCE_METRIC = "Cosine"
|
| 42 |
+
|
| 43 |
+
# Retrieval Configuration
|
| 44 |
+
TOP_K = 5
|
| 45 |
+
SIMILARITY_THRESHOLD = 0.7
|
| 46 |
+
|
| 47 |
+
# RAG Prompt Template
|
| 48 |
+
RAG_PROMPT_TEMPLATE = """You are an expert insurance advisor helping users understand their insurance documents.
|
| 49 |
+
Use the following context from insurance documents to answer the user's question.
|
| 50 |
+
Provide clear, accurate information and explain insurance terms in simple language.
|
| 51 |
+
|
| 52 |
+
Context from insurance documents:
|
| 53 |
+
{context}
|
| 54 |
+
|
| 55 |
+
User Question: {question}
|
| 56 |
+
|
| 57 |
+
Instructions:
|
| 58 |
+
- Answer based primarily on the provided context
|
| 59 |
+
- If the context doesn't contain enough information, clearly state that
|
| 60 |
+
- Explain insurance jargon in plain language
|
| 61 |
+
- For add-on recommendations, compare coverage, costs, and benefits
|
| 62 |
+
- Highlight important exclusions or limitations
|
| 63 |
+
- Be precise about coverage details and conditions
|
| 64 |
+
|
| 65 |
+
Answer:"""
|
| 66 |
+
|
| 67 |
+
# System Instructions for Gemini
|
| 68 |
+
SYSTEM_INSTRUCTION = """You are an expert insurance advisor. Your role is to:
|
| 69 |
+
1. Help users understand complex insurance documents
|
| 70 |
+
2. Explain insurance terms and conditions in simple language
|
| 71 |
+
3. Provide recommendations for add-ons based on coverage gaps
|
| 72 |
+
4. Compare different coverage options clearly
|
| 73 |
+
5. Highlight important exclusions and limitations
|
| 74 |
+
6. Always prioritize accuracy and clarity
|
| 75 |
+
|
| 76 |
+
Remember: You provide informational guidance only, not professional insurance advice."""
|
| 77 |
+
|
| 78 |
+
@classmethod
|
| 79 |
+
def validate_config(cls):
|
| 80 |
+
"""Validate that all required configuration is present"""
|
| 81 |
+
required_keys = [
|
| 82 |
+
("GEMINI_API_KEY", cls.GEMINI_API_KEY),
|
| 83 |
+
("QDRANT_API_KEY", cls.QDRANT_API_KEY),
|
| 84 |
+
("QDRANT_URL", cls.QDRANT_URL)
|
| 85 |
+
]
|
| 86 |
+
|
| 87 |
+
missing = [key for key, value in required_keys if not value]
|
| 88 |
+
|
| 89 |
+
if missing:
|
| 90 |
+
raise ValueError(
|
| 91 |
+
f"Missing required environment variables: {', '.join(missing)}\n"
|
| 92 |
+
"Please set them in your .env file"
|
| 93 |
+
)
|
| 94 |
+
|
| 95 |
+
return True
|
| 96 |
+
|
| 97 |
+
@classmethod
|
| 98 |
+
def get_qdrant_config(cls):
|
| 99 |
+
"""Returns Qdrant configuration dictionary"""
|
| 100 |
+
return {
|
| 101 |
+
"url": cls.QDRANT_URL,
|
| 102 |
+
"api_key": cls.QDRANT_API_KEY,
|
| 103 |
+
"collection_name": cls.COLLECTION_NAME,
|
| 104 |
+
"vector_size": cls.VECTOR_SIZE,
|
| 105 |
+
"distance": cls.DISTANCE_METRIC
|
| 106 |
+
}
|
| 107 |
+
|
| 108 |
+
@classmethod
|
| 109 |
+
def get_chunking_config(cls):
|
| 110 |
+
"""Returns chunking configuration dictionary"""
|
| 111 |
+
return {
|
| 112 |
+
"chunk_size": cls.CHUNK_SIZE,
|
| 113 |
+
"chunk_overlap": cls.CHUNK_OVERLAP,
|
| 114 |
+
"separators": cls.SEPARATORS
|
| 115 |
+
}
|
| 116 |
+
|
| 117 |
+
@classmethod
|
| 118 |
+
def get_retrieval_config(cls):
|
| 119 |
+
"""Returns retrieval configuration dictionary"""
|
| 120 |
+
return {
|
| 121 |
+
"top_k": cls.TOP_K,
|
| 122 |
+
"similarity_threshold": cls.SIMILARITY_THRESHOLD
|
| 123 |
+
}
|