Chatbot_with_Pytorch

A High-Performance Deep Learning Conversational AI Assistant Powered by PyTorch & NLP

---

📑 Table of Contents

• 🌟 Project Overview
• 🧠 Deep Learning & NLP Course Curriculum
• 🔬 Natural Language Processing (NLP) Pipeline
• 🏛️ PyTorch Deep Neural Network Architecture
• ✨ Key Advancements & Features
• 📁 Repository File Tree
• 🚀 Quick Start Installation
• 🎯 Training the Model
• 💬 Running the Chatbot (CLI & Desktop GUI)
• ⚙️ Customizing & Adding New Domains
• 👤 Author & License

---

🌟 Project Overview

Chatbot_with_Pytorch is an enterprise-ready, educational, and high-performance conversational AI system developed and architected by Rahul Chaube.

Unlike naive rule-based keyword matchers, this chatbot utilizes a Deep Multi-Layer Perceptron (MLP) implemented in PyTorch to perform statistical intent classification on tokenized and stemmed text vectors generated via NLTK.

The repository includes:

1. Academic Deep Learning Course (LESSONS.md): Full mathematical derivations, backpropagation formulas, activation proofs, and optimization theory.
2. Interactive Terminal CLI (chat.py): ANSI-colored console interface with slash commands (/help, /tags, /info, /verbose).
3. Desktop Graphical Application (app.py): Modern, dark-themed GUI built with Tkinter.
4. Configurable Training Pipeline (train.py): Hyperparameter CLI flags (--epochs, --lr, --batch-size, --hidden-size, --dropout, --seed).

---

🧠 Deep Learning & NLP Course Curriculum

Explore the full in-depth textbook and course guide in LESSONS.md:

Module	Topic	Core Concepts Covered
Module 1	NLP Foundations	Word Tokenization, Porter Stemmer algorithm, Bag-of-Words (BoW) vector representation
Module 2	Neural Network Architecture	Multi-Layer Perceptrons, Linear transforms, ReLU activations, Dropout regularization, Softmax
Module 3	Training & Optimization	Cross-Entropy Loss mathematical formulation, Adam optimizer moment estimation, mini-batch gradient descent
Module 4	Inference & Production	Dynamic confidence thresholding, Out-of-Distribution (OOD) fallbacks, Tkinter GUI engineering
Module 5	Practical Labs	Hands-on exercises to build custom enterprise chatbots

---

🔬 Natural Language Processing (NLP) Pipeline

The transformation from raw human dialogue to mathematical input tensors proceeds through 4 well-defined phases:

1. Text Tokenization: Splits input strings into discrete linguistic tokens using nltk.word_tokenize.
2. Porter Stemming: Normalizes inflected word variants ("running", "runs", "ran") into their root lemma ("run").
3. Vocabulary Indexing & Bag-of-Words Vectorization: Matches stems against the sorted unique corpus vocabulary $V \in \mathbb{R}^{124}$, producing a binary ${0, 1}$ feature vector.
4. Tensor Conversion: Casts vectors into PyTorch torch.float32 tensors ready for matrix multiplication on CPU or CUDA GPU.

---

🏛️ PyTorch Deep Neural Network Architecture

The core classification engine is a 3-layer Feed-Forward Multi-Layer Perceptron (NeuralNet in model.py):

$$\mathbf{h}_1 = \text{ReLU}(\mathbf{W}_1 \vec{x} + \mathbf{b}_1)$$

$$\mathbf{h}_2 = \text{ReLU}(\mathbf{W}_2 \mathbf{h}_1 + \mathbf{b}_2)$$

$$\mathbf{z} = \mathbf{W}_3 \mathbf{h}_2 + \mathbf{b}_3$$

$$\hat{\mathbf{y}} = \text{Softmax}(\mathbf{z})$$

• Input Dimension ($d_{\text{in}}$): $124$ features (Bag-of-Words vocabulary size).
• Hidden Dimensions ($h_1, h_2$): $16$ units with non-linear ReLU activations and optional Dropout ($p = 0.05$).
• Output Classes ($C$): $13$ intent classes with Softmax probability distribution.

---

✨ Key Advancements & Features

• ⚡ Near-Zero Convergence: Optimized training loop achieving final loss $\le 0.000002$.
• 🛡️ Confidence Threshold Gating: Out-of-domain and low-confidence inputs are intercepted to prevent hallucinated answers.
• 🎨 Modern Minimalist Visuals: High-resolution, clean vector architecture diagrams.
• 🖥️ Cross-Platform Compatibility: Fully tested on Windows (PowerShell/CMD), macOS, and Linux.

---

📁 Repository File Tree

Chatbot_with_Pytorch/
├── assets/
│   ├── banner.png                     # Clean minimalist project hero banner
│   ├── nlp_pipeline.png               # Infographic: Tokenization to BoW tensor
│   └── neural_network_architecture.png# Infographic: PyTorch MLP forward pass
├── intents.json                       # Knowledge base with 13 intent categories
├── model.py                           # PyTorch NeuralNet architecture class
├── nltk_utils.py                      # Tokenization, Stemming & BoW vectorizer
├── train.py                           # CLI-configurable PyTorch training pipeline
├── chat.py                            # Interactive ANSI terminal CLI assistant
├── app.py                             # Modern Tkinter desktop GUI application
├── LESSONS.md                         # Complete deep learning & NLP coursebook
├── data.pth                           # Serialized model checkpoint & vocabulary
├── LICENSE                            # MIT License (Rahul Chaube)
└── README.md                          # Main repository documentation

---

🚀 Quick Start Installation

1. Clone the Repository

git clone https://github.com/Rahulchaube1/Chatbot_with_Pytorch.git
cd Chatbot_with_Pytorch

2. Set Up Virtual Environment

Using uv (Recommended / Ultra-Fast):

uv venv .venv
.\.venv\Scripts\activate
uv pip install torch nltk numpy

Using standard pip:

python -m venv .venv
.\.venv\Scripts\activate
pip install torch nltk numpy

3. Initialize Tokenizer Data (One-Time)

python -c "import nltk; nltk.download('punkt'); nltk.download('punkt_tab')"

---

🎯 Training the Model

Execute the training script to train the neural network on intents.json:

python train.py

Custom Hyperparameter Tuning:

python train.py --epochs 1000 --batch-size 8 --lr 0.001 --hidden-size 16 --dropout 0.05

Flag	Type	Default	Description
--intents	str	intents.json	Path to JSON intent dataset
--output	str	data.pth	Target output model checkpoint file
--epochs	int	1000	Total training epochs
--batch-size	int	8	Mini-batch sample size
--lr	float	0.001	Learning rate for Adam optimizer
--hidden-size	int	16	Number of neurons per hidden layer
--dropout	float	0.05	Dropout regularization probability
--seed	int	42	Random seed for deterministic reproducibility

---

💬 Running the Chatbot

Mode 1: Interactive Terminal CLI

python chat.py --verbose

Supported Terminal Commands:

• /help — Display command list
• /tags — View all 13 recognizable intent tags
• /info — View active PyTorch model architecture and parameters
• /verbose — Toggle live intent and confidence percentages
• quit / exit — Close chat session

---

Mode 2: Modern Desktop GUI Application

python app.py

Launches a standalone desktop application window featuring auto-scrolling dialogue, color-coded speech bubbles, and instant response generation.

---

⚙️ Customizing & Adding New Domains

To train the chatbot for customer support, medical inquiries, e-commerce, or personal portfolios:

1. Open intents.json.
2. Append a new intent block:

{
  "tag": "contact_info",
  "patterns": [
    "How can I reach Rahul Chaube?",
    "Where can I contact the developer?",
    "Show me contact details"
  ],
  "responses": [
    "You can connect with Rahul Chaube via GitHub at https://github.com/Rahulchaube1!"
  ]
}

3. Run python train.py.
4. Your bot is immediately upgraded with the new capabilities!

---

👤 Author & License

• Architect & Lead Developer: Rahul Chaube
• Repository: https://github.com/Rahulchaube1/Chatbot_with_Pytorch
• License: Licensed under the MIT License.