Update app.py

app.py

@@ -7,7 +7,24 @@ from langchain.docstore.document import Document
 from langchain.chains import RetrievalQA
 from langchain_huggingface import HuggingFacePipeline
 from langchain_huggingface.embeddings import HuggingFaceEmbeddings
+from sentence_transformers import SentenceTransformer
+import os
+import pinecone
+import numpy as np
+from langchain.vectorstores import Pinecone
+from langchain.schema import Document
 
+# Initialize Pinecone
+PINECONE_API_KEY = os.getenv("PINECONE_API_KEY", "")
+PINECONE_INDEX = "arolchatbot"  # e.g., "us-west1-gcp-free"
+
+# Connect to Pinecone
+pinecone.init(api_key=PINECONE_API_KEY)
+
+index = pinecone.Index(INDEX_NAME)
+embedder = SentenceTransformer('thenlper/gte-large')
+vector_store = Pinecone(index, embedder.embed_query, "text")
+embeddings = HuggingFaceEmbeddings(model=sentence_model)
 # Model and Tokenizer
 model_name = "Meldashti/chatbot"
 base_model = AutoModelForCausalLM.from_pretrained("unsloth/Llama-3.2-3B")
@@ -28,22 +45,16 @@ generator = pipeline(
 # LLM wrapper
 llm = HuggingFacePipeline(pipeline=generator)
 
-# Embeddings
-embeddings = HuggingFaceEmbeddings(model_name="paraphrase-MiniLM-L3-v2")
-
-# Sample documents (minimal)
-documents = [
-    Document(page_content="Example document about food industry caps"),
-    Document(page_content="Information about manufacturing processes")
-]
+# Wrap the Pinecone index with LangChain's Pinecone wrapper
+vector_store = Pinecone(index, embeddings.embed_query, "text")
 
 # Text splitting
-text_splitter = CharacterTextSplitter(chunk_size=100, chunk_overlap=20)
+text_splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=20)
 split_documents = text_splitter.split_documents(documents)
 
 # Vector store
 vector_store = FAISS.from_documents(split_documents, embeddings)
-retriever = vector_store.as_retriever(search_kwargs={"k": 2})
+retriever = vector_store.as_retriever(search_kwargs={"k": 5})
 
 # Retrieval QA Chain
 rag_chain = RetrievalQA.from_chain_type(