COVID-19 Helpline Chatbot

About the project

A small intent-based chatbot that answers COVID-19 questions — prevention, symptoms, testing, vaccines, and travel — through a desktop chat window. Built from scratch using a trained neural network that classifies user queries into intents and replies from curated response sets, organized as a modular, GitHub-ready codebase (src/, data/, models/, demos/).

What I built / how it worked

• Intent model — a Keras Sequential network (Dense(128, ReLU) → Dropout → Dense(64, ReLU) → Dropout → softmax) classifying 7 intents: greeting, prevention, symptoms, testing, vaccines, travel, thankyou.
• Optimization — trained with SGD (Nesterov accelerated gradient, learning rate 0.01, momentum 0.9) and categorical cross-entropy over 200 epochs.
• Preprocessing — training phrases tokenized and lemmatized with the NLTK WordNet lemmatizer, converted into bag-of-words vectors.
• Interface — a Tkinter desktop GUI titled "COVID-19 CHATBOT SERVICE" with a chat log, message entry box, and send button.
• Packaging — training/inference/UI modules plus serialized word/class stores and model weights (models/words.pkl, classes.pkl, chatbot_model.h5) so the app runs without retraining.

Feature highlights

• End-to-end chatbot from corpus to conversational desktop GUI.
• Neural intent classification (Keras Sequential) with bag-of-words + WordNet lemmatization.
• Predictions matched above a 0.25 error threshold, with the bot replying from the matched intent's response set.