LangChain RAG Project | Part 2: Chunking, FAISS Vector Store & LLM Integration | Video #47 скачать в хорошем качестве

LangChain RAG Project | Part 2: Chunking, FAISS Vector Store & LLM Integration | Video #47

Video #47: RAG Project Part 2: Chunking, Vector Stores & Final Implementation 🚀 Follow the Full Course Playlist here:    • LangChain Full Course: Step-by-Step Tutori...   We are completing the puzzle! 🧩 In Video #47 of our LangChain Full Course, we take the YouTube transcript from the previous video and turn it into a fully functional RAG-based AI Assistant. In this practical, we cover the critical "middle steps" of AI engineering: splitting long text into manageable chunks, converting them into high-dimensional vectors with OpenAI, and storing them in a high-performance FAISS vector store. Finally, we build a custom prompt that forces the LLM to answer only from your provided context, effectively eliminating hallucinations. ✅ In this video, we cover: Recursive Document Chunking: Why chunk_size=500 is the "sweet spot" for transcripts. FAISS Vector Store: Implementing a lightning-fast local vector database. OpenAI Embeddings: Using the text-embedding-3-small model for cost-effective semantic search. Context Augmentation: How to join retrieved documents into a single block of knowledge. Prompt Engineering: Writing a strict system prompt to ensure the AI stays "grounded" in the transcript. The Final Demo: Testing the system with a specific question about the video content. Full Project Code from youtube_transcript_api import YouTubeTranscriptApi from langchain.text_splitter import RecursiveCharacterTextSplitter from langchain_community.vectorstores import FAISS from langchain_openai import OpenAIEmbeddings, ChatOpenAI from langchain_core.prompts import PromptTemplate from dotenv import load_dotenv load_dotenv() 1. Extraction (Recap) ytt = YouTubeTranscriptApi() transcript_obj = ytt.fetch(video_id="qEfPBt9dU60", languages=['en']) subtitles = [snippet.text for snippet in transcript_obj] finaltranscript = " ".join(subtitles) 2. Document Chunking splitter = RecursiveCharacterTextSplitter(chunk_size=500, chunk_overlap=50) listofDocs = splitter.create_documents(texts=[finaltranscript]) 3. Vector Store & Embeddings embeddings = OpenAIEmbeddings(model='text-embedding-3-small') vector_store = FAISS.from_documents(documents=listofDocs, embedding=embeddings) retriever = vector_store.as_retriever() 4. Retrieval question = 'What will happen to the astronauts who are watching the explosion?' results = retriever.invoke(question) doclist = [doc.page_content for doc in results] AugmentedText = "

".join(doclist) 5. LLM Generation with Grounded Prompt llm = ChatOpenAI(model='gpt-3.5-turbo') template = PromptTemplate(template=""" You are a helpful Assistant. Answer from the provided transcript context only. Very Important: If context is insufficient, just say it is not mentioned in the provided context. Following is my context: {context} Question: {question} """, input_variables=['context', 'question']) prompt = template.invoke({'context': AugmentedText, 'question': question}) result = llm.invoke(prompt) print("--- AI RESPONSE ---") print(result.content) #LangChain #RAG #FAISS #OpenAI #AIProject #VectorStore #PromptEngineering #SemanticSearch #LLM #PythonAI #GenerativeAI #CodingTutorial #AIEngineering