import os import io import base64 from PIL import Image as PImage from transformers import pipeline import gradio as gr from dotenv import load_dotenv import requests import json import warnings # --- 1. CONFIGURATION AND CLIENTS --- # Define API endpoints and models (Assuming API key will be handled externally or via environment) warnings.filterwarnings("ignore", category=FutureWarning) # In a Gradio/Hugging Face Space environment, environment variables are typically set in the Space Secrets. # For local testing, ensure your .env has GOOGLE_API_KEY. load_dotenv() google_api_key = os.getenv("GOOGLE_API_KEY") or "" # Assume empty string if not found for Canvas usage if not google_api_key: warnings.warn("GOOGLE_API_KEY not found in environment. Agent features may not work outside of Canvas/properly configured environment.") API_URL = f"https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash-preview-05-20:generateContent?key={google_api_key}" # --- 2. CORE FINANCIAL DATA --- # Your specific details provided in the prompt SALARY = 25000.00 FIXED_EXPENSES = { "SIP": 3000.00, "Mutual_Fund": 1000.00, # Note: MF is usually investment, but treated as a fixed deduction here "Health_Insurance": 2000.00, "Rent/Other_Major_Bills": 12000.00, # Estimated major bills to reach a realistic spending level } # Variable Expenses (Based on your notes) # 3 days in office/week * 4 weeks = 12 trips/month OFFICE_TRIPS_PER_MONTH = 12.0 OFFICE_TRAVEL_COST_AVG = (200 + 400) / 2 # Avg. 300 per round trip OFFICE_FOOD_COST = 180.00 OFFICE_EXPENSE_MONTHLY = OFFICE_TRIPS_PER_MONTH * (OFFICE_TRAVEL_COST_AVG + OFFICE_FOOD_COST) # Weekend Social Expenses (Estimate 2 weekends a month for friends flat visits) WEEKENDS_AT_FRIENDS_PER_MONTH = 2.0 MIDNIGHT_PIZZA_COST = 150.00 COLD_DRINKS_DAILY_COST = 4 * 40 # 4 cold drinks * 40 Rs = 160 per day WEEKEND_EXPENSE_MONTHLY = WEEKENDS_AT_FRIENDS_PER_MONTH * (MIDNIGHT_PIZZA_COST + (2 * COLD_DRINKS_DAILY_COST)) # Assuming 2 days per weekend trip # --- 3. FINANCIAL CALCULATIONS --- def calculate_snapshot(): """Calculates fixed, estimated variable, and net cashflow.""" fixed_total = sum(FIXED_EXPENSES.values()) # Calculate variable totals variable_monthly = OFFICE_EXPENSE_MONTHLY + WEEKEND_EXPENSE_MONTHLY # Total outflow total_outflow = fixed_total + variable_monthly # Net monthly cashflow (before extra spending/movies/shopping) net_cashflow = SALARY - total_outflow snapshot = f""" --- Financial Summary --- Salary (Inflow): ₹{SALARY:,.2f} Fixed Commitments (Outflow): ₹{fixed_total:,.2f} - SIP: ₹{FIXED_EXPENSES['SIP']:,.2f} - Mutual Fund: ₹{FIXED_EXPENSES['Mutual_Fund']:,.2f} - Health Insurance: ₹{FIXED_EXPENSES['Health_Insurance']:,.2f} - Estimated Bills: ₹{FIXED_EXPENSES['Rent/Other_Major_Bills']:,.2f} Estimated Variable Expenses: ₹{variable_monthly:,.2f} - Office Travel/Food (12 days): ₹{OFFICE_EXPENSE_MONTHLY:,.2f} - Weekend Social (2 trips): ₹{WEEKEND_EXPENSE_MONTHLY:,.2f} -------------------------------------- Total Outflow (Fixed + Variable): ₹{total_outflow:,.2f} Net Cashflow (Target Surplus): ₹{net_cashflow:,.2f} -------------------------------------- Goal Note: Your net cashflow should be used for discretionary spending, but you mentioned you often overspend by selling stocks. This is the amount you need to manage carefully. """ return snapshot, net_cashflow # --- 4. GEMINI AGENT LOGIC (Reusing your class structure) --- class GeminiClient: def __init__(self, api_url, financial_data): self.api_url = api_url self.model_name = "gemini-2.5-flash-preview-05-20" self.financial_data = financial_data self.base_system_instruction = f""" You are a personalized Financial Health Coach. Your goal is to provide supportive, actionable, and specific advice based on the user's situation. The user's core financial data is: {financial_data} The user has a major issue with discretionary spending (movies, shopping, extra food delivery) which causes them to sell investments. Always tailor your advice to help the user manage their discretionary spending (their Net Cashflow amount) effectively. If the query requires outside knowledge (like "should I switch my SIP to a different fund?"), use the Google Search tool. """ def chat(self, user_prompt, enable_search=True): payload = { "contents": [{"parts": [{"text": user_prompt}]}], "systemInstruction": {"parts": [{"text": self.base_system_instruction}]}, "config": {} } if enable_search: payload["tools"] = [{"google_search": {}}] try: response = requests.post(self.api_url, headers={'Content-Type': 'application/json'}, data=json.dumps(payload)) response.raise_for_status() result = response.json() candidate = result.get('candidates', [{}])[0] # Extract Text text = candidate.get('content', {}).get('parts', [{}])[0].get('text', "Sorry, the AI could not generate a response.") # Extract Sources sources = candidate.get('groundingMetadata', {}).get('groundingAttributions', []) source_list = [] if sources: source_list = [f"[{i+1}] {s.get('web', {}).get('title', 'Source')} ({s.get('web', {}).get('uri', 'No URI')})" for i, s in enumerate(sources) if s.get('web')] if source_list: text += "\n\n---\n**Sources:**\n" + "\n".join(source_list) return text except Exception as e: return f"An error occurred while connecting to the Gemini API: {e}. Please check your API Key." # Initialize the client with the calculated snapshot initial_snapshot, net_cashflow_val = calculate_snapshot() gemini_client = GeminiClient(API_URL, initial_snapshot) # --- 5. GRADIO INTERFACE --- with gr.Blocks(title="Personal Finance AI Agent") as demo: gr.Markdown("# 💸 Personalized Finance Agent") gr.Markdown("---") # Display the static financial summary first with gr.Row(): with gr.Column(scale=1): gr.Markdown("## Your Financial Snapshot") snapshot_output = gr.Textbox( value=initial_snapshot, label="Monthly Cash Flow Breakdown", lines=15, interactive=False, elem_classes=["font-mono"] ) with gr.Column(scale=1): gr.Markdown("## Key Insight") gr.Markdown(f""" Your current Net Cashflow (Disposable Income before 'extra' spending) is: # **₹{net_cashflow_val:,.2f}** You have been overspending this amount, forcing you to sell investments. This AI Agent will help you manage this discretionary money better. """) # What-if Simulation Tool (Simple calculation for immediate value) gr.Markdown("## What-If Simulator") spending_reduction = gr.Slider( minimum=0, maximum=1000, step=100, value=200, label="Reduce Extra Spending (Zomato/Shopping/Movies) by (₹)", ) def simulate_savings(reduction): annual_savings = reduction * 12 new_net_cashflow = net_cashflow_val + reduction return f""" If you reduce discretionary spending by ₹{reduction:,.0f} per month: - **New Monthly Surplus:** ₹{new_net_cashflow:,.2f} - **Annual Savings Potential:** ₹{annual_savings:,.2f} Ask the AI to help you find ways to save this amount! """ simulation_output = gr.Textbox(label="Simulation Result", lines=6, interactive=False) spending_reduction.change(simulate_savings, inputs=[spending_reduction], outputs=simulation_output) # Initial simulation run simulation_output.value = simulate_savings(200) gr.Markdown("---") # AI Chat Interface gr.Markdown("## 🧠 Personalized Financial Coach Agent") def chat_fn(q, history): # Flatten history for better context passing if needed, but for simplicity, we focus on the current question # Add loading indicator history = history or [] # Determine if search is needed (e.g., asking about external topics like markets or other funds) needs_search = any(word in q.lower() for word in ["stock", "market", "fund", "tax", "switch", "best way", "latest news"]) answer = gemini_client.chat(q, enable_search=needs_search) history.append([q, answer]) return history, gr.Textbox(value="") # Clear user input box chatbot = gr.Chatbot(label="Finance Coach Conversation") user_input = gr.Textbox(label="Ask your Financial Coach (e.g., 'Give me 3 practical ways to save 2000 from my discretionary money,' or 'Should I shift my SIP?', or 'What is a good way to track my spending?')" , placeholder="Ask a question...") submit_btn = gr.Button("Get Advice") submit_btn.click(chat_fn, inputs=[user_input, chatbot], outputs=[chatbot, user_input]) user_input.submit(chat_fn, inputs=[user_input, chatbot], outputs=[chatbot, user_input]) demo.launch()