use gemini-embedding-model

main.py

@@ -5,7 +5,8 @@ from dotenv import load_dotenv
 from utils.uploadFilePDFtoMD import convert_pdf_to_md
 from utils.vectorDB import create_retriever, load_retriever
 from utils.chunking import split_text_by_markdown
-from langchain_community.embeddings import HuggingFaceEmbeddings
+# from langchain_community.embeddings import HuggingFaceEmbeddings
+from langchain_google_genai import GoogleGenerativeAIEmbeddings
 from utils.llm import ask_question
 from pydantic import BaseModel
 
@@ -23,7 +24,7 @@ app.add_middleware(
     allow_headers=["*"],
 )
 
-embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/paraphrase-MiniLM-L12-v2")
+embeddings = GoogleGenerativeAIEmbeddings(model="gemini-embedding-001", api_key=os.getenv("GOOGLE_API_KEY"))
 
 @app.post("/uploadfile/")
 async def upload_file(file: UploadFile = File(...)):
@@ -41,6 +42,7 @@ async def upload_file(file: UploadFile = File(...)):
         md = convert_pdf_to_md(temp_path)
         chunks = split_text_by_markdown(md)
         retriever = create_retriever(chunks, embeddings)
+        # Clean up temp file
         os.remove(temp_path)
         return {"message": "File processed and vector store created successfully."}
     except Exception as e:

utils/vectorDB.py

@@ -1,6 +1,7 @@
 from langchain_pinecone import PineconeVectorStore
 from pinecone import Pinecone, ServerlessSpec
-from langchain.embeddings import HuggingFaceEmbeddings
+from google import genai
+from langchain.embeddings.base import Embeddings
 import os
 from dotenv import load_dotenv
 
@@ -14,18 +15,27 @@ index_name = "rag-chatbot"  # Matches your dashboard name
 if index_name not in pc.list_indexes().names():
     pc.create_index(
         name=index_name,
-        dimension=384,  # MiniLM dims
+        dimension=3072,
         metric="cosine",
         spec=ServerlessSpec(cloud="aws", region="us-east-1")
     )
 
 def create_retriever(chunks, embeddings):
-    pc.Index(index_name).delete(delete_all=True)
+    # Connect to the index
+    index = pc.Index(index_name)
+
+    # Get index stats to check for existing namespaces/vectors
+    stats = index.describe_index_stats()
+
+    # If there are any namespaces (indicating vectors exist somewhere), delete all in the default namespace
+    if 'namespaces' in stats and len(stats['namespaces']) > 0:
+        index.delete(delete_all=True, namespace="")
+
     vector_store = PineconeVectorStore.from_documents(
-        chunks, embeddings, index_name=index_name
+        chunks, embeddings, index_name=index_name, namespace=""
     )
     return vector_store.as_retriever(search_type="similarity", search_kwargs={"k": 5})
 
 def load_retriever(embeddings):
-    vector_store = PineconeVectorStore.from_existing_index(index_name, embeddings)
+    vector_store = PineconeVectorStore.from_existing_index(index_name, embeddings, namespace="")
     return vector_store.as_retriever(search_type="similarity", search_kwargs={"k": 5})