Spaces:
Running
Running
File size: 10,587 Bytes
8e453ef dff2a93 8e453ef 7e9a861 8e453ef a2b02a5 7e9a861 457ebe6 7e9a861 457ebe6 0ec83d3 7e9a861 a2b02a5 7e9a861 a2b02a5 7e9a861 a2b02a5 7e9a861 8e453ef 2497a16 8e453ef 5ed1779 8e453ef |
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 |
# process_aware_rag.py
import os
# Disable telemetry before importing chromadb
os.environ.setdefault("POSTHOG_DISABLED", "true")
os.environ.setdefault("CHROMA_TELEMETRY_DISABLED", "true")
import chromadb
from chromadb.config import Settings
from chromadb.utils import embedding_functions
import google.generativeai as genai
from query_classifier import QueryClassifier
from graph_builder import LegalProcessGraph
from typing import Dict, List, Any
class ProcessAwareRAG:
def __init__(self):
# Initialize components
self.classifier = QueryClassifier()
self.legal_graph = LegalProcessGraph()
self.legal_graph.load_graph('legal_processes.pkl')
# Initialize vector store (use writable path by default)
chroma_path = os.getenv('CHROMA_DB_PATH', '/tmp/legal_vector_db')
os.makedirs(chroma_path, exist_ok=True)
# Redirect model caches to writable directories
default_cache_root = os.getenv('CACHE_ROOT', '/data/cache')
os.environ.setdefault('HOME', '/data')
os.makedirs(default_cache_root, exist_ok=True)
os.makedirs(os.path.join(os.environ['HOME'], '.cache'), exist_ok=True)
os.makedirs(os.path.join(os.environ['HOME'], '.cache', 'chroma'), exist_ok=True)
os.environ.setdefault('HF_HOME', os.path.join(default_cache_root, 'hf'))
os.environ.setdefault('TRANSFORMERS_CACHE', os.path.join(default_cache_root, 'transformers'))
os.environ.setdefault('SENTENCE_TRANSFORMERS_HOME', os.path.join(default_cache_root, 'sentence-transformers'))
os.environ.setdefault('XDG_CACHE_HOME', default_cache_root)
for env_key in ['HF_HOME', 'TRANSFORMERS_CACHE', 'SENTENCE_TRANSFORMERS_HOME', 'XDG_CACHE_HOME']:
os.makedirs(os.environ[env_key], exist_ok=True)
# Disable Chroma anonymized telemetry and initialize client
client = chromadb.PersistentClient(
path=chroma_path,
settings=Settings(anonymized_telemetry=False)
)
# Ensure collection exists
# Use explicit embedding function to ensure queries can compute embeddings
embedding_function = embedding_functions.DefaultEmbeddingFunction()
try:
self.vector_collection = client.get_collection(
"legal_context",
embedding_function=embedding_function
)
except Exception:
self.vector_collection = client.create_collection(
"legal_context",
embedding_function=embedding_function
)
# Initialize LLM
genai.configure(api_key=os.getenv('GOOGLE_API_KEY'))
self.llm = genai.GenerativeModel('gemini-2.5-flash-lite')
def retrieve_graph_context(self, classification: Dict) -> Dict:
"""Retrieve relevant context from knowledge graph"""
graph_context = {
'current_step': None,
'next_steps': [],
'resources': [],
'process_overview': None
}
if classification['process'] == 'general':
return graph_context
process_name = classification['process_name']
# Find current step or process start
if classification['step']:
current_step_id = classification['step']
else:
current_step_id = self.legal_graph.find_process_start(process_name)
if current_step_id:
# Get current step info
if current_step_id in self.legal_graph.graph.nodes:
graph_context['current_step'] = {
'id': current_step_id,
**self.legal_graph.graph.nodes[current_step_id]
}
# Get next steps
graph_context['next_steps'] = self.legal_graph.get_next_steps(current_step_id)
# Get relevant resources
graph_context['resources'] = self.legal_graph.get_required_resources(current_step_id)
return graph_context
def retrieve_vector_context(self, user_query: str, classification: Dict) -> List[Dict]:
"""Retrieve relevant context from vector store"""
# Query vector store
results = self.vector_collection.query(
query_texts=[user_query],
n_results=3,
where={"process": classification.get('process_name', '')} if classification.get('process_name') else None
)
vector_context = []
if results['documents']:
for i in range(len(results['documents'][0])):
vector_context.append({
'content': results['documents'][0][i],
'metadata': results['metadatas'][0][i],
'distance': results['distances'][0][i] if 'distances' in results else None
})
return vector_context
def generate_response(self, user_query: str, graph_context: Dict, vector_context: List[Dict], classification: Dict) -> str:
"""Generate comprehensive response using LLM"""
# Build structured prompt
system_prompt = """
You are a helpful, empathetic, and precise legal guide for Indian legal processes.
IMPORTANT GUIDELINES:
- You must NEVER give legal advice, only guide users through official processes
- Always stress that users should consult a qualified lawyer for their specific case
- Provide specific, actionable steps when possible
- Include official links, phone numbers, and government portals when available
- Be empathetic and understanding of the user's situation
- Use clear, simple language avoiding legal jargon
- Give In The User Language What User Is Communicating or Giving Questions SO Answer In That Language Accordingly
- And Must be Short and Clear Not To give Like long Long Para answers Answer must be Short and Clear
- Dont Give Large Answer Give Answer In One or two three Lines answers Give Accordingly
"""
# Prepare context information
context_sections = []
# Add graph context
if graph_context['current_step']:
context_sections.append(f"""
CURRENT PROCESS STEP:
Title: {graph_context['current_step']['title']}
Description: {graph_context['current_step']['description']}
Properties: {graph_context['current_step'].get('properties', {})}
""")
if graph_context['next_steps']:
next_steps_text = "\n".join([
f"- {step['title']}: {step['description']}"
for step in graph_context['next_steps']
])
context_sections.append(f"""
NEXT STEPS:
{next_steps_text}
""")
if graph_context['resources']:
resources_text = "\n".join([
f"- {res['title']} ({res['type']}): {res['properties'].get('url', res['properties'].get('phone', 'Contact available'))}"
for res in graph_context['resources']
])
context_sections.append(f"""
RELEVANT RESOURCES:
{resources_text}
""")
# Add vector context
if vector_context:
vector_text = "\n\n".join([doc['content'] for doc in vector_context])
context_sections.append(f"""
ADDITIONAL CONTEXT:
{vector_text}
""")
# Build final prompt
full_prompt = f"""
{system_prompt}
USER QUERY: "{user_query}"
CLASSIFICATION: Process: {classification.get('process_name', 'General')}, Intent: {classification.get('intent', 'information')}
{chr(10).join(context_sections)}
Please provide a helpful, structured response that:
1. Acknowledges the user's situation empathetically
2. Provides specific next steps if this is a process guidance request
3. Includes relevant official links and contact information
4. Reminds the user to consult a lawyer for specific legal advice
5. Uses bullet points, bold formatting, and clear structure
Format your response with clear sections and actionable information.
"""
try:
response = self.llm.generate_content(full_prompt)
return response.text
except Exception as e:
return f"I apologize, but I'm having trouble generating a response right now. Please try again or contact NALSA directly at nalsa-dla@nic.in for legal aid queries. Error: {str(e)}"
def process_query(self, user_query: str) -> Dict[str, Any]:
"""Main pipeline: process user query end-to-end"""
# Step 1: Classify query
classification = self.classifier.classify_query(user_query)
# Step 2: Retrieve graph context
graph_context = self.retrieve_graph_context(classification)
# Step 3: Retrieve vector context
vector_context = self.retrieve_vector_context(user_query, classification)
# Step 4: Generate response
response = self.generate_response(user_query, graph_context, vector_context, classification)
return {
'response': response,
'classification': classification,
'graph_context': graph_context,
'vector_context': vector_context,
'debug_info': {
'graph_nodes_found': len(graph_context.get('next_steps', [])),
'vector_docs_found': len(vector_context),
'process_identified': classification.get('process_name')
}
}
# Test the complete pipeline
if __name__ == "__main__":
rag_system = ProcessAwareRAG()
test_query = "I need free legal help but I'm not sure if I qualify. My monthly income is around 45000 rupees."
result = rag_system.process_query(test_query)
print("=== QUERY ===")
print(test_query)
print("\n=== RESPONSE ===")
print(result['response'])
print("\n=== DEBUG INFO ===")
print(f"Process: {result['debug_info']['process_identified']}")
print(f"Graph nodes: {result['debug_info']['graph_nodes_found']}")
print(f"Vector docs: {result['debug_info']['vector_docs_found']}")
|