from langchain.text_splitter import RecursiveCharacterTextSplitter from langchain_huggingface import HuggingFaceEmbeddings from langchain_community.vectorstores import FAISS splitter = RecursiveCharacterTextSplitter(chunk_size=900, chunk_overlap=100) def build_vector_store(transcript: str, cache_folder: str, model_name: str): # cache_folder is ignored for Hugging Face Spaces docs = splitter.create_documents([transcript]) embeddings = HuggingFaceEmbeddings(model_name=model_name) return FAISS.from_documents(docs, embeddings) def get_retriever(vector_store, k: int = 4): return vector_store.as_retriever(search_type="similarity", search_kwargs={"k": k})