vectorstore.py

"""
BƯỚC 4: VECTORSTORE (FAISS in-memory)
-------------------------------------
Tạo FAISS index từ các CHUNK văn bản.
- Không ghi file .faiss nào, tất cả nằm trong RAM.
- Embeddings được lấy từ get_embeddings() (Bước 3).
"""

from langchain_community.vectorstores import FAISS
from embeddings import get_embeddings

def build_vectorstore(chunks):
    """
    Nhận danh sách Document (đã split) và trả về FAISS VectorStore.
    """
    print(">>> Initialising embedding model for FAISS index ...")
    embeddings = get_embeddings()

    print(f">>> Building FAISS index from {len(chunks)} chunks ...")
    vs = FAISS.from_documents(chunks, embeddings)
    print(">>> FAISS index built.\n")
    return vs

if __name__ == "__main__":
    # Test toàn pipeline: load -> split -> FAISS -> similarity_search
    from load_documents import load_documents
    from split_documents import split_documents

    print("=== TEST: load_documents -> split_documents -> FAISS.similarity_search ===\n")

    # 1) Load tài liệu (PDF + HTML) từ HuggingFace
    docs = load_documents()

    # 2) Split thành chunks
    from pprint import pprint
    print(f"Loaded {len(docs)} raw documents.")
    chunks = split_documents(docs)
    print(f"Split into {len(chunks)} chunks.\n")

    # 3) Xây FAISS vectorstore
    vectorstore = build_vectorstore(chunks)

    # 4) Test similarity_search
    query = "Fristen für die Prüfungsanmeldung im Bachelorstudium"
    print("Test query:")
    print(" ", query, "\n")

    results = vectorstore.similarity_search(query, k=3)

    print("Top-3 ähnliche Chunks aus dem VectorStore:")
    for i, doc in enumerate(results, start=1):
        print(f"\n=== RESULT {i} ===")
        print(doc.page_content[:400], "...")
        print("Metadata:", doc.metadata)