app.py

import os
import sqlite3
import requests
import openai
import gradio as gr
import asyncio
from gtts import gTTS
from typing_extensions import TypedDict
from langgraph.graph import StateGraph, START, END
import csv

openai.api_key = os.getenv("OPENAI_API_KEY")

def init_db_from_csv(csv_path: str = "transactions.csv") -> None:
    conn = sqlite3.connect("shop.db")
    cur = conn.cursor()
    cur.execute(
        "CREATE TABLE IF NOT EXISTS transactions (date TEXT, product TEXT, amount REAL)"
    )
    with open(csv_path, newline='') as f:
        reader = csv.DictReader(f)
        rows = [(row["date"], row["product"], float(row["amount"])) for row in reader]
    cur.execute("DELETE FROM transactions")
    cur.executemany(
        "INSERT INTO transactions (date, product, amount) VALUES (?, ?, ?)", rows
    )
    conn.commit()
    conn.close()

init_db_from_csv()

def db_agent(query: str) -> str:
    try:
        conn = sqlite3.connect("shop.db")
        cur = conn.cursor()
        cur.execute(
            """
            SELECT product, SUM(amount) AS revenue
            FROM transactions
            WHERE date = date('now')
            GROUP BY product
            ORDER BY revenue DESC
            LIMIT 1
            """
        )
        row = cur.fetchone()
        if row:
            return f"Top product today: {row[0]} with ₹{row[1]:,.2f}"
        return "No transactions found for today."
    except sqlite3.OperationalError as e:
        return f"Database error: {e}. Please check 'transactions' table in shop.db."

def web_search_agent(query: str) -> str:
    try:
        resp = requests.get(
            "https://serpapi.com/search",
            params={"q": query, "api_key": os.getenv("SERPAPI_KEY")}  
        )
        snippet = resp.json().get("organic_results", [{}])[0].get("snippet", "").strip()
        if snippet:
            return llm_agent(f"Summarize: {snippet}")
    except Exception:
        pass
    return llm_agent(query)

def llm_agent(query: str) -> str:
    response = openai.chat.completions.create(
        model="gpt-4o-mini",
        messages=[
            {"role": "system", "content": "You are a helpful assistant."},
            {"role": "user", "content": query},
        ],
        temperature=0.2,
    )
    return response.choices[0].message.content.strip()

def stt_agent(audio_path: str) -> str:
    with open(audio_path, "rb") as afile:
        transcript = openai.audio.transcriptions.create(
            model="whisper-1",
            file=afile
        )
    return transcript.text.strip()

def tts_agent(text: str, lang: str = 'en') -> str:
    tts = gTTS(text=text, lang=lang)
    out_path = "response_audio.mp3"
    tts.save(out_path)
    return out_path

class State(TypedDict):
    query: str
    result: str

def route_fn(state: State) -> str:
    q = state["query"].lower()
    if any(k in q for k in ["max revenue", "revenue"]):
        return "db"
    if any(k in q for k in ["who", "what", "when", "where"]):
        return "web"
    return "llm"

def router_node(state: State) -> dict:
    return {"query": state["query"]}

def db_node(state: State) -> dict:
    return {"result": db_agent(state["query"]) }

def web_node(state: State) -> dict:
    return {"result": web_search_agent(state["query"]) }

def llm_node(state: State) -> dict:
    return {"result": llm_agent(state["query"]) }

builder = StateGraph(State)
builder.add_node("router", router_node)
builder.set_entry_point("router")
builder.set_conditional_entry_point(
    route_fn,
    path_map={"db": "db", "web": "web", "llm": "llm"}
)
builder.add_node("db", db_node)
builder.add_node("web", web_node)
builder.add_node("llm", llm_node)
builder.add_edge(START, "router")
builder.add_edge("db", END)
builder.add_edge("web", END)
builder.add_edge("llm", END)
graph = builder.compile()

def handle_query(audio_or_text: str):
    is_audio = audio_or_text.endswith('.wav') or audio_or_text.endswith('.mp3')
    if is_audio:
        query = stt_agent(audio_or_text)
    else:
        query = audio_or_text

    state = graph.invoke({"query": query})
    response = state["result"]

    if is_audio:
        audio_path = tts_agent(response)
        return response, audio_path
    return response

with gr.Blocks() as demo:
    gr.Markdown("## Shop Voice-Box Assistant (Speech In/Out)")
    inp = gr.Audio(sources=["microphone"], type="filepath", label="Speak or type your question or upload transactions.csv separately in root")
    out_text = gr.Textbox(label="Answer (text)")
    out_audio = gr.Audio(label="Answer (speech)")
    submit = gr.Button("Submit")
    gr.Examples(
        examples=[
            ["What is the max revenue product today?"],
            ["Who invented the light bulb?"],
            ["Tell me a joke about cats."],
        ],
        inputs=inp,
        outputs=[out_text, out_audio],
    )
    submit.click(fn=handle_query, inputs=inp, outputs=[out_text, out_audio])

if __name__ == "__main__":
    demo.launch(share=False, server_name="0.0.0.0", server_port=7860)