💳 Financial AI System – Complaint Classification & AI Agent

👩‍💻 Overview

This project is an end-to-end AI system for financial text analysis and intelligent automation.

It combines:

• NLP-based classification of financial complaints
• Machine Learning pipeline (TF-IDF + Linear SVM)
• Data engineering and balancing of large-scale datasets
• AI Agent layer for structured decision-making and logging

---

⚙️ Project Structure

1. Complaint Classification Model

• Processes real-world financial complaint data
• Performs text cleaning and preprocessing
• Balances dataset across financial categories
• Trains a supervised ML model (Linear SVM)
• Evaluates performance using classification metrics

2. AI Agent System

• Implements a structured AI agent workflow
• Logs decision-making steps in real time
• Uses modular components:
  • Text cleaning engine
  • Prediction engine
  • Activity logger
• Designed for extensibility and production-style architecture

---

🧠 Machine Learning Pipeline

• Text Vectorization: TF-IDF (1–2 grams)
• Model: Linear Support Vector Machine (LinearSVC)
• Strategy: Stratified train/test split
• Optimization: Balanced class weights
• Output: Multi-class financial category classification

---

📊 Dataset

• Source: Consumer Financial Complaints Dataset
• Large-scale real-world financial narratives
• Processed and balanced across categories:
  • Banking & Cards
  • Debt Collection
  • Loans & Mortgages
  • Credit Services

---

🧹 Preprocessing Steps

• Removal of anonymization patterns
• Text normalization (lowercasing)
• Punctuation removal
• Noise filtering
• Token standardization

---

🤖 AI Agent Layer

The system includes an AI Agent architecture that:

• Tracks execution steps with logging
• Processes input through modular components
• Loads trained ML models dynamically
• Provides structured inference flow

---

📈 Results

• High performance across multiple financial categories
• Robust classification on real-world noisy data
• Stable generalization using SVM with TF-IDF features

---

🛠 Tech Stack

• Python
• Pandas / NumPy
• Scikit-learn
• Matplotlib / Seaborn
• HuggingFace (agent layer)
• Joblib

---

🚀 How to Run

pip install -r requirements.txt
jupyter notebook

Run notebooks in order:

1. Financial_Complaints_Classification_Modeling.ipynb

2. Financial_AI_Agent_Deployment.ipynb

📌 Highlights

• End-to-end ML pipeline
• Real-world financial dataset
• AI agent architecture
• Production-style modular design

📫 Author

Yael Shraga Computer Science Student | Systems & AI Developer

---