Faiaz Amin Khan

I research and build systems that span the physical-digital divide. From designing patent-grade acoustic sensing mechanism at Samsung R&D to deploying AI workflows in resource contrained device, my work focuses on decoding human signals to create seamless, intelligent interactions.

Overview

A lightweight intent-based chatbot for answering university course questions. Uses a bag-of-words pipeline with NLTK preprocessing and a Keras feed-forward network trained on intents.json, plus a simple CLI loop for interaction.

Repo: github.com/faiazamin/ClassBot

Why?

Students often ask the same course questions repeatedly. I wanted a small, editable chatbot that teams can retrain by updating a single intents file—no complex pipelines or external services.

Problem

  • FAQ-style bots are brittle without an easy way to edit intents/responses.
  • Many toy bots ship without retraining scripts, making updates manual.
  • Preprocessing and model artifacts must stay in sync to avoid runtime errors.

Approach

  • Data-driven intents: Patterns and responses live in intents.json.
  • Preprocessing: NLTK tokenization + lemmatization → bag-of-words vectors.
  • Model: Keras/TensorFlow dense network trained via trainingData.py; artifacts (chatbotmodel.h5, words.pkl, classes.pkl) feed the CLI runner.
  • CLI: main.py loads artifacts and runs an interactive loop with exit keywords (bye, Goodbye).

Key Features

  • Editable intents file (intents.json) with bundled examples.
  • Training script to regenerate model + vocab/class pickles.
  • Interactive CLI chatbot for quick demos.
  • Clean repo layout: main.py, trainingData.py, intents.json, saved artifacts.

Tech Stack

  • Python 3.9+
  • TensorFlow / Keras
  • NLTK for tokenization and lemmatization
  • NumPy

My Contributions

  • Built the preprocessing pipeline (tokenize, lemmatize, bag-of-words).
  • Implemented the Keras feed-forward classifier and artifact saving/loading.
  • Authored the training script and interactive CLI runner.
  • Documented setup, NLTK data requirements, and retraining workflow.

Setup & Run

  • Install deps: pip install tensorflow nltk numpy
  • Download NLTK data: python -m nltk.downloader punkt wordnet
  • Run chatbot: python main.py
  • Retrain after editing intents: 
    python trainingData.py

Impact

  • Delivers a modifiable FAQ bot that teams can refresh by editing one JSON file.
  • Keeps preprocessing, model, and artifacts aligned to reduce runtime errors.

Updated: