app.py

import os
import streamlit as st
import pandas as pd
import openai
import torch
import matplotlib.pyplot as plt
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
from dotenv import load_dotenv
import anthropic
import ast
import re

# Load environment variables
load_dotenv()
os.environ["OPENAI_API_KEY"] = os.getenv("OPENAI_API_KEY")
os.environ["ANTHROPIC_API_KEY"] = os.getenv("ANTHROPIC_API_KEY")

# UI Styling
st.markdown(
    """
    <style>
    .stButton button {
        background-color: #1F6FEB;
        color: white;
        border-radius: 8px;
        border: none;
        padding: 10px 20px;
        font-weight: bold;
    }
    .stButton button:hover {
        background-color: #1A4FC5;
    }
    .stTextInput > div > input {
        border: 1px solid #30363D;
        background-color: #161B22;
        color: #C9D1D9;
        border-radius: 6px;
        padding: 10px;
    }
    .stFileUploader > div {
        border: 2px dashed #30363D;
        background-color: #161B22;
        color: #C9D1D9;
        border-radius: 6px;
        padding: 10px;
    }
    .response-box {
        background-color: #161B22;
        padding: 10px;
        border-radius: 6px;
        margin-bottom: 10px;
        color: #FFFFFF;
    }
    </style>
    """,
    unsafe_allow_html=True
)

st.title("Excel Q&A Chatbot 📊")

# Model Selection
model_choice = st.selectbox("Select LLM Model", ["OpenAI GPT-3.5", "Claude 3 Haiku", "Mistral-7B"])

# Load appropriate model based on selection
if model_choice == "Mistral-7B":
    model_name = "mistralai/Mistral-7B-Instruct"
    tokenizer = AutoTokenizer.from_pretrained(model_name)
    model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.float16)
    def ask_mistral(query):
        inputs = tokenizer(query, return_tensors="pt").to("cuda")
        output = model.generate(**inputs)
        return tokenizer.decode(output[0])

elif model_choice == "Claude 3 Haiku":
    client = anthropic.Anthropic(api_key=os.environ["ANTHROPIC_API_KEY"])
    def ask_claude(query):
        response = client.messages.create(
            model="claude-3-haiku",
            max_tokens=512,
            messages=[{"role": "user", "content": query}]
        )
        return response.content[0]["text"]

else:
    client = openai.OpenAI()
    def ask_gpt(query):
        response = client.chat.completions.create(
            model="gpt-3.5-turbo",
            messages=[{"role": "user", "content": query}]
        )
        return response.choices[0].message.content

# File Upload with validation
uploaded_file = st.file_uploader("Upload a file", type=["csv", "xlsx", "xls", "json", "tsv"])

if uploaded_file is not None:
    file_extension = uploaded_file.name.split(".")[-1].lower()
    
    try:
        if file_extension == "csv":
            df = pd.read_csv(uploaded_file)
        elif file_extension in ["xlsx", "xls"]:
            df = pd.read_excel(uploaded_file, engine="openpyxl")
        elif file_extension == "json":
            df = pd.read_json(uploaded_file)
        elif file_extension == "tsv":
            df = pd.read_csv(uploaded_file, sep="\t")
        else:
            st.error("Unsupported file format. Please upload a CSV, Excel, JSON, or TSV file.")
            st.stop()
        
        st.write("### Preview of Data:")
        st.write(df.head())
        
        # Extract metadata
        column_names = df.columns.tolist()
        data_types = df.dtypes.apply(lambda x: x.name).to_dict()
        missing_values = df.isnull().sum().to_dict()
        
        # Display metadata
        st.write("### Column Details:")
        st.write(pd.DataFrame({"Column": column_names, "Type": data_types.values(), "Missing Values": missing_values.values()}))
        
    except Exception as e:
        st.error(f"Error loading file: {str(e)}")
        st.stop()
    
    # User Query
    query = st.text_input("Ask a question about this data:")
    
    if st.button("Submit Query"):
        if query:
            # Interpret the query using selected LLM
            if model_choice == "Mistral-7B":
                parsed_query = ask_mistral(f"Convert this question into a structured data operation: {query}")
            elif model_choice == "Claude 3 Haiku":
                parsed_query = ask_claude(f"Convert this question into a structured data operation: {query}")
            else:
                parsed_query = ask_gpt(f"Convert this question into a structured data operation: {query}")
            
            # Validate and clean query
            parsed_query = re.sub(r"[^a-zA-Z0-9_()\[\]"'., ]", "", parsed_query.strip())
            st.write(f"Parsed Query: `{parsed_query}`")
            
            # Predefined Safe Execution Methods
            SAFE_OPERATIONS = {
                "sum": lambda col: df[col].sum(),
                "mean": lambda col: df[col].mean(),
                "max": lambda col: df[col].max(),
                "groupby_sum": lambda col, group_by: df.groupby(group_by)[col].sum()
            }
            
            # Safe Execution
            try:
                exec_result = eval(parsed_query, {"df": df, "pd": pd, "SAFE_OPERATIONS": SAFE_OPERATIONS})
                st.write("### Result:")
                st.write(exec_result if isinstance(exec_result, pd.DataFrame) else str(exec_result))
                
                # If numerical data, show a visualization dynamically
                if isinstance(exec_result, pd.Series):
                    fig, ax = plt.subplots()
                    if exec_result.dtype in ["int64", "float64"]:
                        exec_result.plot(kind="bar", ax=ax)
                    elif exec_result.dtype == "object":
                        exec_result.value_counts().plot(kind="bar", ax=ax)
                    st.pyplot(fig)
                
            except SyntaxError as e:
                st.error(f"Syntax Error in parsed query: {str(e)}")
            except Exception as e:
                st.error(f"Error executing query: {str(e)}")
    
    # Memory for context retention
    if "query_history" not in st.session_state:
        st.session_state.query_history = []
    st.session_state.query_history.append(query)