app.py

import gradio as gr
import numpy as np
import re
from datetime import datetime, timedelta
import difflib
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch
import yfinance as yf  # Added for dynamic data fetching

# Hardcoded common symbols for fuzzy matching (expand as needed)
available_symbols = ['TSLA', 'MSFT', 'NVDA', 'GOOG', 'AMZN', 'SPY']

# Financial calculation functions (updated for dynamic data)
def parse_period(query):
    match = re.search(r'(\d+)\s*(year|month|week|day)s?', query.lower())
    if match:
        num = int(match.group(1))
        unit = match.group(2)
        if unit == 'year':
            return timedelta(days=365 * num)
        elif unit == 'month':
            return timedelta(days=30 * num)
        elif unit == 'week':
            return timedelta(weeks=num)
        elif unit == 'day':
            return timedelta(days=num)
    return timedelta(days=365)  # Default to 1 year

def find_closest_symbol(input_symbol):
    input_symbol = input_symbol.upper()
    closest = difflib.get_close_matches(input_symbol, available_symbols, n=1, cutoff=0.6)
    return closest[0] if closest else None

def calculate_growth_rate(start_date, end_date, symbol):
    try:
        ticker = yf.Ticker(symbol)
        hist = ticker.history(start=start_date, end=end_date)
        if hist.empty:
            return None
        beginning_value = hist.iloc[0]['Close']  # Use 'Close' if 'Adj Close' not available; adjust as needed
        ending_value = hist.iloc[-1]['Close']
        dividends = hist['Dividends'].sum() if 'Dividends' in hist.columns else 0
        total_return = (ending_value - beginning_value + dividends) / beginning_value
        years = (end_date - start_date).days / 365.25
        if years == 0:
            return 0
        cagr = (1 + total_return) ** (1 / years) - 1
        return cagr * 100  # As percentage
    except Exception as e:
        print(f"Error fetching data for {symbol}: {e}")
        return None

def calculate_investment(principal, years, annual_return=0.07):
    return principal * (1 + annual_return) ** years

# Load SmolLM3 model and tokenizer
model_name = "HuggingFaceTB/SmolLM3-3B"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    device_map="auto",  # Auto-detect GPU/CPU
    torch_dtype=torch.bfloat16 if torch.cuda.is_available() else torch.float32,
    # Optional: load_in_8bit=True for quantization (requires bitsandbytes)
)

def generate_response(user_query, enable_thinking=True):
    # Parse query for symbol and period
    symbol_match = re.search(r'\b([A-Z]{1,5})\b', user_query.upper())
    symbol = find_closest_symbol(symbol_match.group(1)) if symbol_match else 'SPY'  # Default to SPY
    period = parse_period(user_query)
    end_date = datetime.now()
    start_date = end_date - period

    # Calculate growth rate if applicable
    growth_rate = calculate_growth_rate(start_date, end_date, symbol)
    if growth_rate is not None:
        summary = f"The CAGR for {symbol} over the period is {growth_rate:.2f}%."
    else:
        summary = f"No data available for {symbol} in the specified period."

    # Handle investment projection
    investment_match = re.search(r'\$(\d+)', user_query)
    if investment_match:
        principal = float(investment_match.group(1))
        years = period.days / 365.25
        projected = calculate_investment(principal, years)
        summary += f" Projecting ${principal} at 7% return over {years:.1f} years: ${projected:.2f}."

    # Prepare prompt for model
    system_prompt = "You are a knowledgeable financial advisor. Provide accurate, concise advice based on the data provided. Disclaimer: This is not professional advice."
    messages = [
        {"role": "system", "content": system_prompt},
        {"role": "user", "content": f"{summary} {user_query}"}
    ]

    text = tokenizer.apply_chat_template(
        messages,
        tokenize=False,
        add_generation_prompt=True,
        enable_thinking=enable_thinking  # Enable deep reasoning for complex queries
    )

    model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
    generated_ids = model.generate(
        **model_inputs,
        max_new_tokens=100,
        temperature=0.6,
        top_p=0.95,
        repetition_penalty=1.2,
        do_sample=True
    )

    output_ids = generated_ids[0][len(model_inputs.input_ids[0]):]
    response = tokenizer.decode(output_ids, skip_special_tokens=True)
    return response.strip()

# Gradio interface
def chat(user_input, history):
    response = generate_response(user_input)
    history.append((user_input, response))
    return history, ""

with gr.Blocks() as demo:
    gr.Markdown("# FinChat: AI-Powered Financial Advisor")
    chatbot = gr.Chatbot()
    msg = gr.Textbox(placeholder="Ask about stocks, investments, etc.")
    clear = gr.Button("Clear")

    msg.submit(chat, [msg, chatbot], [chatbot, msg])
    clear.click(lambda: None, None, chatbot, queue=False)

demo.launch()