diff --git a/backend/app/core/lifespan.py b/backend/app/core/lifespan.py
new file mode 100644
index 0000000..7f3df4b
--- /dev/null
+++ b/backend/app/core/lifespan.py
@@ -0,0 +1,50 @@
+from contextlib import asynccontextmanager
+
+import sqlalchemy
+from fastapi import FastAPI
+from llama_index.vector_stores.postgres import PGVectorStore
+from openai import AsyncOpenAI
+
+from app.core.logging import logger
+
+from .config import settings
+from .db import init_db
+
+
+class AppState:
+    groq_client: AsyncOpenAI | None = None
+    vector_client: PGVectorStore | None = None
+
+
+app_state = AppState()
+
+
+@asynccontextmanager
+async def lifespan(app: FastAPI):
+
+    try:
+        init_db()
+        logger.info("Database initialized")
+
+        app_state.groq_client = AsyncOpenAI(
+            base_url=settings.LLM_ENDPOINT,
+            api_key=settings.LLM_API_KEY or "missing-key",
+        )
+
+        url = sqlalchemy.make_url(settings.DATABASE_URL)
+        app_state.vector_client = PGVectorStore.from_params(
+            host=url.host,
+            port=str(url.port or 5432),
+            user=url.username,
+            password=url.password,
+            database=url.database,
+            table_name="regulations_vectors",
+            embed_dim=1024,
+        )
+        yield
+
+    finally:
+        if app_state.groq_client:
+            await app_state.groq_client.close()
+        if app_state.vector_client:
+            await app_state.vector_client.close()
diff --git a/backend/app/regintel/rag.py b/backend/app/regintel/rag.py
index 049e2e2..cf1f249 100644
--- a/backend/app/regintel/rag.py
+++ b/backend/app/regintel/rag.py
@@ -1,28 +1,20 @@
-from functools import lru_cache
 from pathlib import Path
 
-import sqlalchemy
 from llama_index.core import SimpleDirectoryReader, VectorStoreIndex
 from llama_index.core.node_parser import SentenceSplitter
 from llama_index.core.schema import NodeWithScore
 from llama_index.embeddings.cohere import CohereEmbedding
 from llama_index.vector_stores.postgres import PGVectorStore
-from openai import AsyncOpenAI
 
 from app.core.config import settings
 
+from ..core.lifespan import app_state
+
 # Detect project root (where .env lives)
 # rag.py is in backend/app/regintel/
 BASE_DIR = Path(__file__).resolve().parent.parent.parent.parent
 
 
-@lru_cache
-def get_openai_client():
-    return AsyncOpenAI(
-        base_url=settings.LLM_ENDPOINT, api_key=settings.LLM_API_KEY or "missing-key"
-    )
-
-
 class VectorStoreConnection:
     def __init__(self):
         self.splitter = SentenceSplitter(chunk_size=512, chunk_overlap=60)
@@ -30,17 +22,9 @@ def __init__(self):
         self.should_reset = False
 
     @property
-    def vector_store(self):
-        url = sqlalchemy.make_url(settings.DATABASE_URL)
-        return PGVectorStore.from_params(
-            host=url.host,
-            port=str(url.port or 5432),
-            user=url.username,
-            password=url.password,
-            database=url.database,
-            table_name="regulations_vectors",
-            embed_dim=1024,
-        )
+    def vector_store(self) -> PGVectorStore:
+        assert app_state.vector_client is not None, "Vector client is not initialized"
+        return app_state.vector_client
 
     @property
     def embedding_model(self):
@@ -89,8 +73,8 @@ async def vector_chat_async(self, query: str):
         {query}
         """
         print(user_prompt)
-
-        completion = await get_openai_client().chat.completions.create(
+        assert app_state.groq_client is not None, "Groq client is not initialized"
+        completion = await app_state.groq_client.chat.completions.create(
             model=settings.CHAT_MODEL,
             messages=[
                 {"role": "system", "content": system_prompt},
diff --git a/backend/main.py b/backend/main.py
index ca1e6e6..b9b44d3 100644
--- a/backend/main.py
+++ b/backend/main.py
@@ -1,22 +1,11 @@
-from contextlib import asynccontextmanager
-
 from fastapi import FastAPI
 from fastapi.middleware.cors import CORSMiddleware
 
 from app.api import activity, ingest, manifest, validate
 from app.core.config import settings
-from app.core.db import init_db
-from app.core.logging import logger
+from app.core.lifespan import lifespan
 from app.core.rate_limit import setup_rate_limiting
 
-
-@asynccontextmanager
-async def lifespan(app: FastAPI):
-    init_db()
-    logger.info("Database initialized")
-    yield
-
-
 app = FastAPI(title=settings.PROJECT_NAME, lifespan=lifespan)
 
 # CORS Configuration