ローカル環境で動作するRAGを構築してみた

Q: How do I reset my password if I can't access my registered email?
A: If you do not have access to your registered email, please contact support. You may be asked to verify your identity with other information.

Q: Can I enable two-factor authentication for added security?
A: Yes, you can enable 2FA from your account security settings, which requires an authentication app such as Google Authenticator or Authy.

Q: What should I do if my account is suspended due to suspicious activity?
A: Contact support immediately. We will guide you through the authentication process to regain access.

Q: What should I do if I accidentally delete a shared file on my company drive?
A: Contact IT Support immediately. In most cases, we can restore files deleted within the last 30 days from our backup system.

Q: Can I install the software on my company laptop myself?
A: No, you are not allowed to install software without IT approval. You must apply for this through the software installation form.

Q: How do I report a suspicious email or phishing attempt?
A: Forward the email to phishing@yourcompany.com and delete it. Do not click on any links or open any attachments.

from pathlib import Path
from llama_index.core import VectorStoreIndex, StorageContext, load_index_from_storage, Document
from llama_index.embeddings.huggingface import HuggingFaceEmbedding
from llama_index.core.settings import Settings
from llama_cpp import Llama

MODEL_PATH = "/llama_models/tinyllama-1.1b-chat-v1.0.Q4_0.gguf"
llm = Llama(model_path=MODEL_PATH, n_ctx=2048)

Settings.llm = None
PERSIST_DIR = "./storage_txt"
EMBED_MODEL = HuggingFaceEmbedding(model_name="/embedding_models/intfloat-multilingual-e5-small")

def load_faq_documents(path: str):
    with open(path, "r", encoding="utf-8") as f:
        content = f.read()
    return [Document(text=entry.strip()) for entry in content.strip().split("\n\n") if entry.strip()]

if not Path(PERSIST_DIR).exists():
    print("📂 Indexing FAQ...")
    documents = load_faq_documents("docs/faq.txt")
    index = VectorStoreIndex.from_documents(documents, embed_model=EMBED_MODEL)
    index.storage_context.persist(persist_dir=PERSIST_DIR)
else:
    print("📦 Loading existing index...")
    storage_context = StorageContext.from_defaults(persist_dir=PERSIST_DIR)
    index = load_index_from_storage(storage_context, embed_model=EMBED_MODEL)

retriever = index.as_retriever(similarity_top_k=1)

print("\n💬 FAQ Search + TinyLlama Generator (English only)")
print("Type 'exit' to quit.")

while True:
    query = input("\n📝 Your question: ")
    if query.lower() == "exit":
        break

    nodes = retriever.retrieve(query)
    context = nodes[0].text if nodes else "No relevant FAQ found."

    print("\n✅ context:")
    print(context)

    prompt = f"""You are a helpful assistant. Given the following context and question, provide a concise and accurate answer.

Context:
{context}

Question:
{query}

Answer:"""

    output = llm(prompt, max_tokens=150, echo=False)
    english_answer = output["choices"][0]["text"].strip()

    print("\n✅ English Answer:")
    print(english_answer)