Portfolio

GitHub Accounts

Data Science

My Datasets

1. Pakistan-PHD country directory (PCD)
   

   A list of all the PhDs Pakistani universities have produced.

2. Ph.Ds. produced by Pakistanis universities 2010-14
   

   Ph.Ds. produced by Pakistan’s public and private sector universities from 2010 to 2014.

3. Districts_of_Pakistan
   

   A dataset detailing the districts of Pakistan.

Deep Learning

Simple CNN models - Happyface and Digit hand Signs

• This repository highlights projects showcasing deep learning applications using TensorFlow and Keras.
• Focuses on convolutional neural networks (ConvNets) for image classification tasks.
• Demonstrates practical use of both the Sequential and Functional APIs for model development.

ResNet - Digit hand Signs

• The Kaggle notebook demonstrate the developement, training, and testing of ResNet Model using TensorFlow and Keras. The validation accuracy of 100% is achieved by best model using checkpoints.

Transfer Learning using MobileNet - alpaca, non-alpaca dataset

• The project demonstrate the developement, training, and testing of MobileNet Model using TensorFlow and Keras through Transfer Learning for alpaca, non-alpaca dataset.

• The kaggle notebook demonstrate 100% accurate MobileNet Model using TensorFlow and Keras through Transfer Learning for alpaca, non-alpaca dataset and advanced data augmentation techniques.

Object Detection using yolov2

• This notebook demonstrates object detection using YOLOv2, leveraging technologies such as TensorFlow for deep learning and the PIL library for image processing to identify and classify objects with bounding boxes for 80 classes supported by the COCO dataset. Users can process any input image, visualize detection results, and benefit from the model’s efficiency and versatility in real-time applications.

Image segmentation using Unet

• This project focuses on developing and optimizing a U-Net model for image segmentation tasks. It involves training on RGB images paired with corresponding masks to accurately segment images.

Face recognition using facenet

• This project implements a state-of-the-art facial recognition system using the FaceNet model, which generates 512-dimensional embeddings of face images to perform verification and recognition tasks.

DL Art - Neural Style Transfer

• This project applies neural style transfer to blend the content of a target image with the artistic style of a reference image using deep learning techniques. By leveraging pre-trained convolutional neural networks (VGG19) for feature extraction and optimizing the generated image through a combination of content and style costs, the project generates visually compelling images that merge both elements seamlessly.

RNN from Scratch - Dinosaur Island

• This project involves developing a character-level Recurrent Neural Network (RNN) from scratch to generate unique dinosaur names by training on a dataset of existing names. By learning the patterns and structures of these names, the model aims to produce innovative and safe new names for fictional dinosaurs in a creative setting.

Text generation - Transfer Learning - LSTM based RNN - Shakespeare Sonnet

• This project uses LSTM based RNN for character-level text generation to emulate Shakespearean-style poetry. By training on a corpus of Shakespeare’s works, the model captures intricate linguistic patterns and dependencies, aiming to produce text that reflects his distinct style and structure.

Music Generation - LSTM based RNN

• This project uses LSTM based RNN for character-level text generation to emulate Shakespearean-style poetry. By training on a corpus of Shakespeare’s works, the model captures intricate linguistic patterns and dependencies, aiming to produce text that reflects his distinct style and structure.

Word Embeddings - Similarity, Debiasing & Equalization

• This project explores the identification and reduction of gender biases in word embeddings, specifically using GloVe vectors, through techniques like neutralization and equalization. It applies these methods to various words and gendered pairs, demonstrating how debiasing can mitigate implicit biases in natural language processing models.

Emojifier: Enhancing Text Expressiveness with Emoji

• This project focuses on predicting emojis based on the sentiment conveyed in a sentence using word embeddings and LSTM networks. The Emojify-V1 Model uses basic word embeddings for emoji prediction, while the Emojify-V2 Model leverages LSTM networks to better understand word order and context, improving accuracy in predicting emojis even in more complex sentences.

Neural Machine Translation with Attention

• This project demonstrates the application of an attention-based neural network to perform sequence-to-sequence translation, using the example of date format conversion. Key features include the attention mechanism, which enables the model to selectively focus on relevant input parts for accurate output generation, and the ability to visualize attention weights, offering insights into the model’s decision-making.

Trigger word detection - from voice

• This project focuses on building a trigger word detection system using deep learning, leveraging synthetic data generation and advanced recurrent neural network architecture with GRUs for robust performance. Key features include spectrogram-based audio preprocessing, chime overlay functionality upon trigger word detection, and a customizable system capable of handling diverse audio inputs and noisy environments.

Transformer from Scratch

• This project implements the Transformer architecture from scratch, encompassing positional encodings, scaled dot-product attention, multi-head attention, and the complete Encoder and Decoder. It provides a detailed understanding of the fundamental components that make up the Transformer model.

Exploring Positional Encodings in Transformer Architectures

• This project investigates the role of positional encodings in Transformer models by visualizing their relational properties and their impact on semantic word embeddings. Key features include correlation and distance matrix analysis, integration with GloVe embeddings, and the ability to adjust weights to balance semantic meaning with positional context.

Transformer Network Application: Named-Entity Recognition (NER)

• This project leverages Transformer models for Named-Entity Recognition (NER), focusing on preprocessing techniques to optimize model performance on tokenized textual data. Key features include aligning true labels with tokenized inputs and visualizing label distributions to assess prediction consistency, achieving high accuracy and F1-scores across various entity categories.

Transformer Network Application: Question Answering

• This project successfully implemented and compared Transformer-based models for extractive Question Answering using Hugging Face Transformers, TensorFlow, and PyTorch, demonstrating effective fine-tuning, preprocessing, and evaluation techniques. It showcased the strengths of each framework and provided insights into applying advanced models for accurate and efficient natural language understanding.

Machine Learning

SpaceX Falcon 9 ML Project

This notebook details an ML project focusing on SpaceX Falcon 9 launches that encompasses:

• Data Collection via API and web scraping
• Data Wrangling and Exploratory Data Analysis (EDA)
• Visualization and Interactive Dashboards using Plotly Dash and Folium
• Predictive Analysis through classification techniques

Exploratory Data Analysis

Tesla and GameStop Stock/Revenue Data and Dashboard

This notebook details an exploratory data analysis on historical stock data focusing on Tesla and GameStop Stock/Revenue Data that encompasses:

• Fetch Data: Utilizing the yfinance library to extract stock data for your company of interest
• Analyze: Explore key metrics, visualize trends, and draw insights from the data.
• Report Findings: Summarize the analysis and trends in alignment with real-world market behavior and financial performance.

Socioeconomic Indicators in Chicago (2008-2012)

This notebook presents a comprehensive exploratory data analysis (EDA) of socioeconomic indicators in Chicago from 2008 to 2012. Through a combination of visualizations and statistical summaries, this notebook aims to uncover trends and insights related to Chicago’s socioeconomic landscape during the specified period. The analysis includes:

• Pairplots: Visualizing relationships between multiple variables.
• Heatmaps: Illustrating data correlation and distributions.
• Correlation Matrix: Examining relationships between different socioeconomic features.
• Descriptive Statistics: Summarizing key metrics of the dataset.
• Detailed Analysis: In-depth exploration of the correlation matrix for various features.

Ecommerce Best Selling Category Analysis

This notebook provides a detailed analysis of Pakistan’s Largest E-Commerce Dataset that encompasses:

• Data Loading: Importing and cleaning the dataset.
• Data Analysis: Analyzing order quantities and visualizing trends by category and year.
• Key Insights: Identifying best-selling categories and understanding data distribution.

Pakistan-PHD Country Directory (PCD) - EDA

This notebook focuses on exploratory data analysis (EDA) of the Pakistan-PHD country directory (PCD), offering:

• PhD Production: Overview of PhD production trends in Pakistan.
• Subject Distribution: Analysis of popular PhD subjects.
• University Performance: Comparison of university performance based on PhD output.

Dashboard & Visualization Projects

Sales and Service Analysis Report for SwiftAuto Traders - Looker Dashboard Project

This Looker report captures the detailed analysis and visualizations for both the Sales and Service dashboards, allowing for a comprehensive view of the performance metrics at SwiftAuto Traders.

This report presents an analysis of car sales and profits for each dealer at SwiftAuto Traders. The analysis aims to provide insights into key performance indicators (KPIs) that can assist in making informed business decisions. The report is divided into two main sections: Sales and Service.

Sales Dashboard

KPI Metrics

• Total Profit: $X.XX million
• Total Quantity Sold: Y units
• Average Quantity Sold: Z units

Visualizations

1. Quantity Sold by Model
2. Profit by Dealer ID

Service Dashboard

Visualizations

1. Number of Recalls per Model
2. Customer Sentiment Analysis
3. Monthly Car Sales vs. Profit
4. Recalls by Model and Affected System

Products and Sales Analysis Report for Customer Loyality Program - Looker Dashboard Project

KPI Metrics

• Total Revenue: $X.XX million
• Total Quantity Sold: Y units

Visualizations

1. Line Chart - Quantity Sold of d/f Product lines by Year
2. Bar Chart - Total Quantity Sold of to Male or female customers - Gender Slicer
3. Line and bar chart for average unit sale price and total revenue with product line slicer.
4. Quantity Sold of d/f product lines on world map
5. heat map of quantity sold on world map.
6. treemap of quantity sold and revenue by country, state, and city.
7. Word cloud of revenue by state or province.
8. Bubble chart of revenue by loyality status and product line.

Blockchain and Smart Contracts

FL-Incentivizer

• Created ERC20 and ERC721 based tokens to incentivize local models training and global model trading respectively.

Whitelist-Dapp

• This DApp whitelists up to 10 addresses for the presale of NFTs.

NFT Collection

• This DApp mints up to 20 NFTs. It allows only whitelisted addresses from the Whitelist-DApp to mint NFTs during the presale period. After the presale ends, anyone can mint publicly.

Basic DApp

• This DApp allows users to set their mood in a smart contract.

ERC20 Based Cryptocurrency

• This project involves creating a fungible token following the ERC-20 standard as a custom cryptocurrency.

Basic NFT Contract

• This project focuses on building a basic NFT (Non-Fungible Token) contract on the Ethereum network using Hardhat and OpenZeppelin Contracts.