bin/rag_system.py

"""
Portable RAG (Retrieval-Augmented Generation) System
=====================================================
A completely portable, offline-capable RAG setup using ChromaDB and FastEmbed.

Usage:
  1. Add a Document:
     python rag_system.py ingest my_document.pdf
     python rag_system.py ingest my_notes.txt

  2. Query the Knowledge Base:
     python rag_system.py query "What is the main topic?"
"""

import sys
import os
import argparse
from pathlib import Path

# Fix for ChromaDB SQLite requirement in some environments
import sqlite3
import chromadb

# LangChain components
from langchain_community.document_loaders import PyPDFLoader, TextLoader
from langchain_text_splitters import RecursiveCharacterTextSplitter
from langchain_community.vectorstores import Chroma

# HuggingFace for Embeddings ^& LLM Generation
from langchain_huggingface import HuggingFaceEndpoint, HuggingFaceEmbeddings
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.runnables import RunnablePassthrough
from langchain_core.output_parsers import StrOutputParser

# Configuration
DEVTOOLS_ROOT = Path(os.environ.get("DEVTOOLS_ROOT", os.path.dirname(os.path.dirname(os.path.abspath(__file__)))))
RAG_DATA_DIR = DEVTOOLS_ROOT / "rag_data"
RAG_DATA_DIR.mkdir(exist_ok=True)


# 1. Initialize Embeddings (LOCAL - No Token Needed)
print("[INFO] Loading Embedding Model (Local - This may take a moment on first run)...")
embeddings = HuggingFaceEmbeddings(
    model_name="sentence-transformers/all-MiniLM-L6-v2",
    model_kwargs={'device': 'cpu'}
)

# 2. Initialize Chroma Vector Database
print(f"[INFO] Connecting to ChromaDB at {RAG_DATA_DIR}...")
vectorstore = Chroma(
    embedding_function=embeddings,
    persist_directory=str(RAG_DATA_DIR)
)


def ingest_document(file_path: str):
    """Reads a file, splits it into chunks, and saves to Vector DB."""
    path = Path(file_path)
    if not path.exists():
        print(f"[ERROR] File not found: {file_path}")
        sys.exit(1)

    print(f"\n[1/3] Loading document: {path.name}...")
    if path.suffix.lower() == '.pdf':
        loader = PyPDFLoader(str(path))
    elif path.suffix.lower() == '.txt':
        loader = TextLoader(str(path), encoding='utf-8')
    else:
        print("[ERROR] Unsupported file type. Please use .pdf or .txt")
        sys.exit(1)

    docs = loader.load()
    print(f"      Loaded {len(docs)} pages/sections.")

    print("[2/3] Splitting into chunks...")
    text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200)
    splits = text_splitter.split_documents(docs)
    print(f"      Created {len(splits)} chunks.")

    print("[3/3] Generating vectors and saving to database...")
    vectorstore.add_documents(splits)
    print("\n[SUCCESS] Document ingested successfully!")


def format_docs(docs):
    return "\n\n".join(doc.page_content for doc in docs)


def query_rag(query: str):
    """Retrieves relevant chunks and optionally uses an LLM to generate an answer."""
    print(f"\n[Q] {query}\n")

    # Step 1: Retrieval
    print("─── RETRIEVED CONTEXT ────────────────────")
    retriever = vectorstore.as_retriever(search_kwargs={"k": 3})
    results = retriever.invoke(query)
    
    if not results:
        print("No relevant context found in the database.")
        return

    for i, doc in enumerate(results):
        source = doc.metadata.get('source', 'Unknown')
        page = doc.metadata.get('page', '')
        page_info = f" (Page {page})" if page else ""
        print(f"[{i+1}] Source: {os.path.basename(source)}{page_info}")
        print(f"    {doc.page_content[:300]}...\n")

    # Step 2: Generation (Optional - Requires Token)
    hf_token = os.environ.get("HF_TOKEN")
    if hf_token:
        print("\n─── AI GENERATED ANSWER ──────────────────")
        try:
            llm = HuggingFaceEndpoint(
                repo_id="meta-llama/Meta-Llama-3-8B-Instruct",
                task="text-generation",
                max_new_tokens=512,
                huggingfacehub_api_token=hf_token
            )
            
            template = """You are an assistant for question-answering tasks. 
Use the following pieces of retrieved context to answer the question. 
If you don't know the answer, say that you don't know. 
Keep the answer concise and accurate based ONLY on the context.

Context: {context}

Question: {question}

Answer:"""
            prompt = ChatPromptTemplate.from_template(template)
            
            rag_chain = (
                {"context": retriever | format_docs, "question": RunnablePassthrough()}
                | prompt
                | llm
                | StrOutputParser()
            )
            
            response = rag_chain.invoke(query)
            print(response.strip())
            
        except Exception as e:
            print(f"[WARN] Could not generate AI answer. (Are you logged into Hugging Face?)")
            print(f"       Error: {e}")
    else:
        print("\n[NOTE] HuggingFace token not found. Only returning retrieved documents.")
        print("       Run 'hf auth login' in your terminal to enable AI Generation.")


def main():
    parser = argparse.ArgumentParser(description="Portable RAG System")
    subparsers = parser.add_subparsers(dest="command", help="Commands")

    # Ingest command
    ingest_parser = subparsers.add_parser("ingest", help="Ingest a document into the database")
    ingest_parser.add_argument("file", type=str, help="Path to PDF or TXT file")

    # Query command
    query_parser = subparsers.add_parser("query", help="Query the database")
    query_parser.add_argument("query", type=str, help="Your question")

    args = parser.parse_args()

    if args.command == "ingest":
        ingest_document(args.file)
    elif args.command == "query":
        query_rag(args.query)
    else:
        parser.print_help()


if __name__ == "__main__":
    main()