agent_backend.py

import os
import sqlite3
from langchain.agents import create_sql_agent, AgentType, initialize_agent
from langchain.tools import Tool
from langchain.llms import Groq
from langchain.sql_database import SQLDatabase

# --------------------------
# Load DB
# --------------------------
DB_PATH = "customer_orders.db"
conn = sqlite3.connect(DB_PATH)

db = SQLDatabase.from_uri(f"sqlite:///{DB_PATH}")

# --------------------------
# Initialize LLM using HF secret
# --------------------------
llm = Groq(
    model="llama-3.3-70b-versatile",
    groq_api_key=os.getenv("GROQ_API_KEY")
)

# --------------------------
# Create SQL agent
# --------------------------
sql_agent = create_sql_agent(
    llm=llm,
    db=db,
    agent_type=AgentType.ZERO_SHOT_REACT_DESCRIPTION,
    verbose=False
)

# --------------------------
# Safety wrapper
# --------------------------
def safe_generate(prompt):
    guardrail = f"""
You are a FoodHub support chatbot.

Rules:
- No hallucinating order details.
- No SQL logs or internal reasoning.
- Answer politely and clearly.

USER:
{prompt}
"""
    try:
        return llm.invoke(guardrail).content
    except:
        return "Sorry, something went wrong."

# --------------------------
# Intent Classifier
# --------------------------
def intent_classifier(query):
    prompt = f"""
Classify the user's intent. Respond with EXACTLY one label:

needs_order_id
sql_query
general_query

User query: {query}
"""
    return llm.invoke(prompt).content.strip().lower()

# --------------------------
# Query Normalizer
# --------------------------
def normalize_query(user_query):
    prompt = f"""
Rewrite the user's message into a short instruction for retrieving order info.
Do NOT write SQL. Only describe what to retrieve.

User:
{user_query}
"""
    try:
        return llm.invoke(prompt).content
    except:
        return user_query

# --------------------------
# Order Query Tool
# --------------------------
def order_query_tool_function(query):
    try:
        cleaned = normalize_query(query)
        result = sql_agent.invoke({"input": cleaned})
        raw = result.get("output", "").strip()

        if not raw or "i don't know" in raw.lower() or "no results" in raw.lower():
            return "The order ID you provided does not exist."

        return raw

    except:
        return "Error retrieving data. Please check the order ID."

order_query_tool = Tool(
    name="OrderQueryTool",
    func=order_query_tool_function,
    description="Fetch raw order details."
)

# --------------------------
# Answer Refiner
# --------------------------
def answer_refiner_tool_function(raw):
    if "does not exist" in raw.lower():
        return raw

    prompt = f"""
Rewrite the raw order data in a simple, polite, customer-friendly way.

RAW:
{raw}
"""
    return safe_generate(prompt)

answer_refiner_tool = Tool(
    name="AnswerRefinerTool",
    func=answer_refiner_tool_function
)

# --------------------------
# Chat Agent (tools + LLM)
# --------------------------
chat_agent = initialize_agent(
    tools=[order_query_tool, answer_refiner_tool],
    llm=llm,
    agent_type=AgentType.ZERO_SHOT_REACT_DESCRIPTION,
    verbose=False
)

# --------------------------
# Smart Router
# --------------------------
def smart_chat_router(user_input):
    intent = intent_classifier(user_input)

    if intent == "needs_order_id":
        return "Could you please provide the order ID?"

    elif intent == "general_query":
        return safe_generate(user_input)

    else:
        result = chat_agent.invoke({"input": user_input})
        output = result.get("output", "")

        if not output or "i don't know" in output.lower():
            return "Sorry, I could not find that order."

        return output

def process_message(msg):
    return smart_chat_router(msg)