Ivory: AI-Powered Financial Assistant

Ivory is our submission for the Tech Solstice Codeathon at MAHE Bangalore on 5th February 2025.

This project aims to simplify personal finance management using ML-based insights and a conversational AI interface powered by Google Gemini. It combines automated financial analysis, personalized nudges, and an interactive chatbot — making smart financial decision-making accessible to everyone.

---

Inspiration

Managing personal finances is often overwhelming — from tracking expenses to budgeting and making sense of spending habits. As students, we saw how many of our peers struggle with this. Ivory was born from this challenge — to create a smart financial assistant that:

• Tracks and analyzes expenses

• Predicts spending behavior

• Offers personalized savings suggestions

• Provides interactive financial advice via AI

---

Tech Stack & Architecture

Ivory is built with a modular structure to ensure scalability and ease of development.

Backend

• Python with Flask for serving API endpoints.

• Gemini API for conversational AI.

• Pandas/Numpy/Scikit-learn for data handling and ML analysis.

• Mockaroo used during development for realistic sample financial datasets.

Core Features

• Financial data processing & feature engineering

• ML models for behavior prediction and savings suggestions

• Categorization and anomaly detection

• Interactive chatbot for AI-powered insights

---

Project Structure

Ivory/
├── backend/
│   ├── financial_behavior_ml2/
│   │   ├── models/
│   │   ├── models_corrected/
│   │   ├── analyze_csv.py
│   │   ├── api_helper.py
│   │   ├── basic_eval.py
│   │   ├── bias_corrected_models.py
│   │   ├── data_processor.py
│   │   ├── evaluate_models.py
│   │   ├── finance_chatbot.py
│   │   ├── main.py
│   │   ├── model_evaluator.py
│   │   ├── model_trainer.py
│   │   ├── nudge_generator.py
│   │   ├── simple_eval.py
│   │   ├── test_gemini_api.py
│   │   ├── financial_behavior.csv
│   │   ├── model_accuracy.csv
│   │   └── model_accuracy_matrix.csv
│   └── requirements.txt
│
├── financial_behaviour_ml/
│   ├── calendar/
│   ├── models/
│   ├── src/
│   │   ├── api.py
│   │   ├── behavior_predictor.py
│   │   ├── data_processor.py
│   │   ├── expense_categorizer.py
│   │   └── savings_optimizer.py
│   ├── tests/
│   │   ├── test_categorizer.py
│   │   ├── test_optimizer.py
│   │   └── test_predictor.py
│   ├── app.py
│   ├── config.py
│   ├── financial_behavior.csv
│   └── requirements.txt
│
├── frontend/
│   ├── page/
│   │   ├── images/
│   │   ├── calendar.html
│   │   ├── index.html
│   │   ├── landing.html / .css / .js
│   │   ├── opt_out.html / .js
│   │   ├── rewards.html / .css
│   │   ├── stats.html / .js
│   │   └── style.css
│
├── models/
│   ├── eating_out_savings_model.joblib
│   ├── education_savings_model.joblib
│   ├── groceries_savings_model.joblib
│   ├── healthcare_savings_model.joblib
│   ├── miscellaneous_savings_model.joblib
│   ├── transport_savings_model.joblib
│   ├── utilities_savings_model.joblib
│   └── entertainment_savings_model.joblib

├── .gitignore
└── README.md

---

Setup Instructions

1. Install Dependencies:

   pip install -r requirements.txt
   

2. Add Financial Dataset: Ensure financial_behavior.csv is placed in the appropriate directory.

3. Set Up Gemini API Key:

   • Get one from Google AI Studio

   • Replace the default key in finance_chatbot.py if needed

---

Usage

Run Financial Analysis:

python main.py

Test Gemini API:

python test_gemini_api.py [YOUR_API_KEY]

Start Chatbot:

python finance_chatbot.py

The chatbot will:

• Run analysis

• Parse ML results

• Answer finance-related queries with personalized insights

---

Sample Questions for Chatbot

• “How can I improve my savings?”

• “What are the biggest drains on my finances?”

• “Can you recommend a budget?”

• “Am I spending too much on food?”

• “How do I optimize transportation costs?”

---

How It Works

1. ML Layer:

   • Models evaluate spending patterns and highlight potential savings.

   • Includes bias correction and savings optimization logic.

2. AI Layer (Gemini):

   • Parses model output

   • Provides contextual, user-specific responses via chat

3. Frontend Layer:

   • HTML/CSS-based dashboard for visualization

   • User-friendly interface for input & recommendations

---

What We Learned

• Handling real-world Flask backend architecture

• Integrating external APIs (Mockaroo, Gemini)

• ML model training & evaluation under time constraints

• Team collaboration with Git under pressure

• Designing scalable, modular systems

---

Challenges Faced

• Merge conflicts during rapid development

• Gemini model changes requiring updates

• Data inconsistencies from mock sources

• Limited time to implement full ML pipeline

---

What’s Next?

• Replace rule-based insights with fully trained ML models

• Add manual entry & opt-out options

• Launch a fully interactive dashboard

• Integrate more advanced NLP techniques for conversation analysis

---