Merge pull request #22 from kamyabnazari/kn-vector-db-swtich-and-fixes

Changing to qdrant db

backend/chat_bot_function.py

@@ -1,36 +1,33 @@
 import os 
-from langchain.vectorstores.chroma import Chroma
 from langchain.chat_models import ChatOpenAI
 from langchain.embeddings.openai import OpenAIEmbeddings
 from langchain.chains import ConversationalRetrievalChain
 from langchain.schema import HumanMessage, AIMessage
+from langchain.vectorstores import Qdrant
+from qdrant_client import QdrantClient
+
 from dotenv import load_dotenv
 
+load_dotenv()
 
 def make_chain(documentId:str):
     model = ChatOpenAI(
         model_name="gpt-3.5-turbo",
         temperature="0",
-        # verbose=True
-    )
-    embedding = OpenAIEmbeddings()
-
-    vector_store = Chroma(
-        collection_name=documentId,
-        embedding_function=embedding,
-        persist_directory= os.getenv('DB_PERSIST_DIRECTORY'),
     )
+
+    embeddings = OpenAIEmbeddings()
 
+    client = QdrantClient(os.getenv('PUBLIC_QDRANT_URL'))
+    qdrant = Qdrant(client, documentId, embeddings)
+
     return ConversationalRetrievalChain.from_llm(
         model,
-        retriever=vector_store.as_retriever(),
+        retriever=qdrant.as_retriever(),
         return_source_documents=True,
-        # verbose=True,
     )
 
-
 def chat_bot_funtion(question: str, chat_history, documentId: str):
-    load_dotenv()
     if not chat_history:
         chat_history=[]
 

backend/ingest.py

@@ -2,16 +2,18 @@
 import PyPDF3
 import re
 import os
-import sys
 from typing import Callable, List, Tuple, Dict
 
 from langchain.docstore.document import Document
 from langchain.text_splitter import RecursiveCharacterTextSplitter
-from langchain.vectorstores import Chroma
 from langchain.embeddings import OpenAIEmbeddings
+from langchain.vectorstores import Qdrant
+
+from qdrant_client import QdrantClient
 
 from dotenv import load_dotenv
 
+load_dotenv
 
 def extract_metadata_from_pdf(file_path: str) -> dict:
     with open(file_path, "rb") as pdf_file:
@@ -108,9 +110,8 @@ def text_to_docs(text: List[str], metadata: Dict[str, str]) -> List[Document]:
 
     return doc_chunks
 
-
 def create_embeddings_from_pdf_file(file_path: str, documentId: str):
-
+    
     # Step 1: Parse PDF
     raw_pages, metadata = parse_pdf(file_path)
 
@@ -128,12 +129,12 @@ def create_embeddings_from_pdf_file(file_path: str, documentId: str):
 
     # Step 3 + 4: Generate embeddings and store them in DB
     embeddings = OpenAIEmbeddings()
-    vector_store = Chroma.from_documents(
+
+    client = QdrantClient(os.getenv('PUBLIC_QDRANT_URL'))
+    qdrant = Qdrant(client, documentId, embeddings)
+
+    qdrant.from_documents(
         document_chunks,
         embeddings,
         collection_name=documentId,
-        persist_directory=os.getenv('DB_PERSIST_DIRECTORY'),
-    )
-
-    # Save DB locally
-    vector_store.persist()
+    )

backend/main.py

@@ -12,7 +12,7 @@
 from fastapi.middleware.cors import CORSMiddleware
 
 # Importing langchain
-from langchain import LLMChain, PromptTemplate
+from langchain import LLMChain
 from langchain.chat_models import ChatOpenAI  
 from langchain.prompts import ChatPromptTemplate
 from langchain.schema import HumanMessage, AIMessage

backend/requirements.txt

@@ -9,7 +9,6 @@ pocketbase
 PyPDF3
 pdfkit
 pytest
-chromadb
 pdfplumber
 qdrant-client
 retry

frontend/README.md

@@ -9,6 +9,7 @@ Create a `.env` file in the root of the frontend directory with the following va
 ```
 PUBLIC_POCKETBASE_URL=http://localhost:8090
 PUBLIC_BACKEND_URL=http://localhost:5003
+PUBLIC_QDRANT_URL=http://localhost:6333
 PUBLIC_SECURE=true
 PUBLIC_HTTPONLY=false
 PUBLIC_SAMESITE=None

frontend/src/lib/components/FileTable.svelte

@@ -6,6 +6,8 @@
 	import type { Record } from 'pocketbase';
 	import { onMount } from 'svelte';
 	import { getDocumentURL } from '$lib/utils';
+	import { env } from '$env/dynamic/public';
+	import axios from 'axios';
 
 	onMount(async () => {
 		await fetchDocuments();
@@ -28,6 +30,11 @@
 	async function deleteDocument(documentID: string) {
 		try {
 			await pb.collection('documents').delete(documentID);
+			await axios({
+				url: `${env.PUBLIC_QDRANT_URL}/collections/${documentID}`,
+				method: 'delete',
+				headers: { 'Content-Type': 'application/json' }
+			});
 			documentList = documentList.filter((document) => document.id !== documentID);
 		} catch (error) {
 			console.error('Fetch error:', error);