BankMate - Smart Banking Bot Assistant

A fast, lightweight, guidance-only banking chatbot
Built for Nigerian banks & fintechs • Fully customizable • No real actions performed

Features

• Guidance-only — explains procedures, never performs real banking actions
• 13 banking intents (balance check, transfer, lost card, loan, bills, airtime, fraud, complaints, USSD, ATM/branch, etc.)
• 682+ unique training examples — realistic Nigerian English + Pidgin
• 12–15 natural responses per intent — feels very human, never robotic
• Ultra-lightweight model (~300–800 KB) • < 5 ms inference • CPU-only
• Easy customization — edit CSV (examples) or JSON (responses) in seconds
• Beautiful modern Gradio UI — very responsive
• Insightful visualizations — class distribution, word clouds, confusion matrix, top features, t-SNE
• Ready for production — FastAPI / Docker templates

Demo

Screenshots

Modern chat UI with typing animation & natural replies

Class distribution & confusion matrix (generated during training)

Quick Start

# 1. Clone the repo
git clone https://github.com/apdoolhamza/BankMate-AI/.git
cd BankMate-AI

# 2. Install dependencies
pip install -r requirements.txt

# 3. Launch the chatbot
python app.py

Customization (No coding required)

Change responses (tone, Pidgin, bank name, etc.)

• Open bankmate_responses.json
• Edit any list of replies
• Save → restart Gradio

Add new intents or more examples

• Open bankmate_intents_data.csv in Excel / Google Sheets
• Add rows like this:

new_intent,how do I reset my password
new_intent,what to do if I forget PIN

• Save file
• Re-run training codes Notebook once

Visualizations (Auto-generated during training)

• Intent distribution bar chart
• Word clouds (per intent)
• Confusion matrix heatmap
• Top contributing features per intent
• t-SNE semantic separation plot All are in /images/ folder — perfect for reports or presentations.

Tech Stack

• Core Model: scikit-learn (TF-IDF + Logistic Regression)
• Interface: Gradio (modern chat UI)
• Data: CSV (examples) + JSON (responses)
• Visualization: Matplotlib, Seaborn, WordCloud, t-SNE
• Model Saving: joblib

License

MIT License – feel free to use, modify, and deploy commercially.

Author

Apdoolmajeed Hamza (apdoolhamza)
AI/ML Engineer | Full-stack Web Developer

• LinkedIn: https://www.linkedin.com/in/apdoolhamza/
• GitHub: https://github.com/apdoolhamza/

Star ⭐ the repo if you find it helpful!