Hola, soy yo!

Qaiser Abbas

I'm an experienced

Experienced Deep Learning and Computer Vision Engineer with 4+ years of industrial experience. Skilled in Deep Learning, Computer Vision, NLP, and Data Science. Strong R&D professional currently working as a Sr. Machine Learning Engineer at Octopus Digital. I am a passionate developer who thrives to design and develop efficient A.I solutons to solve industrial problems.

I got my bachelors from the department of CS & IT, University of Sargodha with a gold medal. My BS was financially supported by University of Sargodha's merit scholarship & Fauji Foundation scholarhip. I obtained my Masters degree in Computer Science (Deep Learning and Medical Image Analysis) from one of the oldest and prestigious engineering universities in Pakistan — University of Engineering and Technology, Lahore.

I have worked in diverse fields of Artificial Intelligence including deep learnig, computer vision, natural language processing and data science. My main area of expertise is in industrial problem solving using state of the art AI tools and technologies. I like to design and develop AI solutions for various business problems involving textual, visual and time series data.

I am learning and practicing MLOps and Cloud technologies for Machine Learning development. I also love to do researach and read about the latest technologies in the field. I regularly follow top notch researchers and labs in the field so I can use thier algorithms in my problems. I am a medium pacer and love to bowl. I am also a casual photographer and a universal foodie.

Research Interests

Software Engineering

Machine Learning

Computer Vision

Natural Language Processing

Computational Neuroscience

Medical Image Analysis

Data Science

Image Processing

MS in Computer Science

University of Engineering and Technology, Lahore

Sep 2018 - Oct 2020

Thesis: Detection and Prediction of Acral Lentiginous Melanoma in Dermoscopic Images using Deep Learning

Summary: Acral melanoma is life threatning cancer. Due to acral melanoma infrequent occurrences, limited data is available so its early diagnosis is hard. To overcome this problem, we applied data centric techniques to develop a large(comparatively) dataset to train a deep learning models. Our proposed convolutional neural network achieved an accuracy of 91% on test set.

Relevant Coursework

Machine Learning
Deep Learning
Computer Vision
Research Methods
Software Quality Assurance

UET Lahore

BS in Information Technology

University of Sargodha, Sargodha

Oct 2014 - May 2018

FYP: IoT based Low Cost Intelligent Surveillance System for Smart Home Security

Summary: I worked on IOT based smart solution for smart home security. We utilized a RaspberryPi and a Pi Camera and PIR motion sensors to develop a prototype of security system. Final system was able to capture suspicious movements and notify the home owner via push and email notifications.

Relevant Coursework

Network Security
Cloud Computing
Artificial Intelligence
Human Computer Interaction
Information Technology Infrastructure

Online Certifications

Computer Vision

State University of New York | Certificate

Deep Learning Specialization

DeepLearning.AI | Certificate

Maths for ML - Linear Algebra

Imperial College London | Certificate

Maths for ML - Multivariate Calculas

Imperial College London | Certificate

An Intuitive Introduction to Probability

University of Zurich | Certificate

AI for Medical Diagnosis

DeepLearning.AI | Certificate

Introduction to Intel® Distribution of OpenVINO™

Intel | Certificate

Python Basics

University of Michigan | Certificate

Python Functions Files Dictionaries

University of Michigan | Certificate

Python Classes and Inheritance

University of Michigan | Certificate

Python Software Engineering and Image Manipulation

University of Michigan | Certificate

English for Career Development

University of Pennsylvania | Certificate

Octopus Digital

August 2024 - Present

Senior Machine Learning Engineer

Applying Machine Learning across Manufacturing, Energy, Oil & Gas, and Consumer Goods sectors.
Focusing on predictive maintenance, industrial fault detection, industrial quality prediction, and demand forecasting.
Developing cutting-edge products that enhance operational efficiency and performance, driving innovation and optimizing pro- cesses in diverse industrial applications.

SDSol Technologies

November 2022 - August 2024

Machine Learning Engineer

Designed and Developed an Audio Analysis Pipeline for Psychological Applications using LLMs (Whisper and GPT).
Developed a GPU-accelerated backend daemon on EC2 for audio and text data processing for therapy sessions notes generation.
Utilized LLMs (GPTs) for development of a personalized treatment plan generation for mental health.
Finetuned GPT 3.5 Turbo for custom treatment plan generation using OpenAI Python SDK and deployed using FastAPI on AWS EC2.
Developed a website assistant chatbot using GPT and LlamaIndex and deployed using FastAPI.
Developed a tennis player/ball detection model via finetuning YOLOv8 and integrated it into active learning pipeline for automatic annotation using LabelStudio
Utilized deep transfer learning for development of aerial scene understanding models using remote sensing images.
Developed a human related crime identification model via finetuning EfficientNet and trained on custom dataset.
Leveraged AWS services including EC2, S3, Lambda, ECR, and AppRunner to ensure seamless and scalable system deployment
Utilized unsupervised association rule mining to craft a personalized recipe recommendation system and implemented it as a robust REST API using FastAPI.
Developed an insect classification model using Transfer Learning, optimized for CPU inference using Intel OpenVino and deployed using ACR and Azure App service.
Developed a Chatbot for question answering for multiple websites and PDF documents using LangChain and Llama-index (Used GPT-4, BERT, LongFormer for generating answers)
Developed an end-to-end query classification pipeline. Collected data from various resources, used ChatGPT API for generating synthetic data and trained Logistic Regression, SVM and Small Language Models (BERT, ALBERT) for query analysis.
Developed a video subtitle generation pipeline using OpenAI's Whisper and deployed as a custom software solution.
Developed an audio to video lip movement synchronization script leveraging a state of the art Wav2Lip Model.
Developed an API for sentence's semantic similarity calculation using Sentence Transformers.
Designed and developed multiple UIs for Machine Learning model demos using Streamlit.
Collaborated with backend developers to deploy and integrate ML models with existing systems in production.

Wortel AI

November 2021 - January 2022

Software Engineer (Deep Learning and Computer Vision)

Worked on Agricultre related projects such as weed detection using YOLOv5 algorithm
Developed a medical speech recognition system by fine‑tuning an Nvidia QuartzNet model via NeMo library
Worked on AWS S3 and MLFlow platform for deployment and maintenance of deep learning models.
Worked on researching various methods for converting deep learning models to work with different ML libraries.

University of Engineering & Technology, Lahore

March 2021 - September 2022

AI Instructor

I was previously assoiated with UET Lahore.
Taught undergraduate AI courses and conducted practical labs.
Delivered practical lectures on Machine Learning and Deep Learning.
Assisted the senior faculty in designing course contents and writing proposals for research grants.
Helped and guided final-year students in their ML based FYPs.
Worked on researching the latest AI technologies to solve real-world problems.

Achievements

A research proposal entitled "Tea disease detection using Machine Learning and Remote Sensing" that I wrote won a grant of PKR 3.5 Million from Higher Education Commission's National Research Program for Universities.
Published a journal paper entitled "Detection and Classification of Malignant Melanoma Using Deep Features of NASNet"

UpWork

November 2020 - March 2021

Freelance Deep Learning Engineer

Developed an image captioning algorithm for image retrieval using image’s natural language description.
Designed and developed a GAN model for Covid detection in CT Scans.
Worked on various object detection projects using YOLO models.

Al-Khawarizmi Institute of Computer Science

January 2020 - October 2020

Research Assistant (Deep Learning)

Worked on Melanoma Diagnosis in Bioinformatics Research Lab at KICS UET Lahore where I used Computer-Based Diagnosis (Deep Learning and Computer Vision) to detect and classify a rare type of skin cancer. Two research papers has been published from the outcomes of this research.
Worked with Prof. Dr. Muhammad Usman Ghani Khan on detection of rare and lethal Acral Lentiginous Melanoma.
Developed a detection system for acral melanoma in dermoscopic images using proposed CNN architecture.
Worked on Plant Disease Detection datasets and deployed classification models as REST APIs
Worked on Intelligent crop disease detection system using deep transfer learning.
Developed a deep learning algorithms for steel defect classification

Wizdojo Technologies

August 2019 - December 2019

Computer Vision Engineer

Assisted in design and development of a Video Analytics System for cars at parking station which stores information related to cars, where it was parked, it's number plate and will inform user whether the specific place is available for parking or not.
Research and develop best vehicle registration plate segmentation model.
Develop a large dataset from videos and annotation of images dataset.
Training, Testing and evaluation of deep learning model (Mask R-CNN) using test dataset.
Research Python code optimization techniques.
Developed a small website using HTML, CSS and Bootstrap.

Al-Khawarizmi Institute of Computer Science

October 2018 - December 2018

Computer Vision Intern

Worked as a Computer Vision and Deep Learning Intern.
Learnt the basics of ANNs, CNNs and deep learning.
Developed small CNNs for object recognition.
Developed GUIs for Object Recognition models deployment (HTML/CSS/Bootstrap).

YouTube Visual Data Collection
This is a small CLI tool to download YouTube frames without downloading and manually extracting frames from YouTube videos. Just provide a YouTube video link and the CLI tool will automatically download the frames in your target directory.
Tech Stack & Libraries : Python, OpenCV, VidGear, ArgrParse
GitHub Link

UOS

Video Subtitles Generation
A python program for generating video subtitles. The program utilizes OpenAI's Whisper model for converting speech to text and then this data is used to generate subtitle files. The program currently can generate .srt and .vtt foramts for subtitles.
Tech Stack & Libraries : Python, whisper
GitHub Link

UOS

DeepDS: A python cli tool for dataset creation
This is a useful cli tool currently under development. It accepts a video (local video file or YouTube link) and utilized optimized deep learning model to generate datasets for high level computer vision tasks. It reads video, runs deep learning model on individual frames and classified and sorts output frames into seprate directory based on predicted class.
Tech Stack & Libraries : Python, OpenVino, Inception V3, VidGear, ArgreParse
Github Link

UOS

Python Automation Scripts
This repo has multiple automation scripts using Python. The scripts can be used in any computer vision projects. It has scripts to read, write, resize and perform multiple operations on images.
Tech Stack & Libraries : Python, OpenCV, Augmentor, PyTorch, NumPy, Pandas
GitHub Link

UOS

Metal Surface Defect Classification
In this project a dataset for steel defect detection by NEU was used classification of steel surface defects. Dataset has total six classes. A ResNet-18 was finetuned to classify the input images into one of the six classes. The model achieved 97% accuracy (approx.)
Tech Stack & Libraries : Python, FastAI, PyTorch, OpenCV, Pandas, Scikit-learn, Matplotlib
GitHub Link

UOS

Detection & Classification of Firearms
This was a term project for my Machine Learning course @UET during my MS. We collected and anotated a large datasets of firearms divided into five categories. Data was collected from YouTube and other video files from web. A YOLOv3 model was trained for detection and classification while a Mask-RCNN model was trained for instance segmentation of firearms. We achieved a 97% accuracy for detection and classification.
Tech Stack & Libraries : Python, PyTorch, FastAI, OpenCV, yt_downloader
GitHub Link | Report

Check out all my contributions @ github/qaixerabbas

UOS

Adversarial Attacks on Video Anomaly Detection Models
We worked on Adversarial Attacks on AIOT Operations Technolgies with a focus on theme parks. In this project we developed a robust video classification model for human related crimes identification using efficient and compute efficient convolutional architectures. To show the impact of adversarial attacks on such life threatning scenarios we used various sparse black box attacks on video classification models. A detailed paper is being written and its in progress.
Tech Stack & Libraries : Python, OpenCV, PyTorch, Keras, TensorFlow, ResNets, MobileNets, EfficienNets, Huggingface, TorchAttacks
GitHub Link | Adv Attack Paper

UOS

Aerial Scene Understanding using Satellite Images
In this project we extensively experimented with three remote sensing image classification datasets using recent convolutional and attention based architectures. The focus was to identify best models keeping compute performance, prediction accuracy and number of parameters in mind. This project helps the newbies and researchers to select the best models when working with aerial scene understanding. This project is under active development and we are experimenting with attention based architectures for now.
Tech Stack & Libraries : Python, OpenCV, PyTorch, ResNets, EfficienNets, ConvNext, DenseNet, SqueezeNet
GitHub Link | Aerial Scene Understanding Paper

UOS

Skin Lesion Analysis for Skin Cancer Detection
This was my MS thesis project. I worked on skin cancer (Acral Melanoma) detection and classification from Dermoscopy images. A data centric approach was used to enhance limited training data. Various deep learning models were trained to select the best model. Our AMNet model was a modified ResNet-18 architecture that performed better to diagnose Acral Melanoma. In addition, a NASNet model was also used for binary classification of Melanoma skin cancer.
Tech Stack & Libraries : Python, OpenCV, Augmentor, PyTorch, Keras, TensorFlow, ResNets, MobileNets, NASNet
GitHub Link | Acral Melanoma Detection | Melanoma Detection

UOS

Punjabi Named Entity Recognition
This project tackles the problem of NER for low resource languages. We worked on data generation for Punjabi (Shahmukhi Script). A multilingual BERT model was trained for NER task. We developed a large training corpora using a simple and novel PoW augmentation technique. The publication is accepted in the prestigious ACM TALLIP.
Tech Stack & Libraries : Python, BERT, NLTK, SpaCy
GitHub Link | Paper

UOS

Efficient Network Intrusion Detection for Secure Systems
We experimented with various machine learning models for recent intrusion detection datasets. NSL-KDD, CSE-CIC-IDS2018 and UNSW-NB-15 datasets were used in the study. Five major ML algorithms were trained on preprocessed datasets. We proposed a hybrid, cost and compute efficient ensemble model for development of secure IDS systems.
Tech Stack & Libraries : Python, scikit-learn, Pandas, NumPy, Keras
GitHub Link | Paper

UOS

Lung Cancer Classification from Limited Data
We used a recent dataset named as LC-25000 for lungs cancer classification. The dataset contains 15000 images of lungs histopathology. We trained two ResNets (18 and 50), MobileNet and AlexNet model for classification. A maximum accuracy of 98% was achived using ResNet-18 algorithm.
Tech Stack & Libraries : Python, PyTorch, FastAI, OpenCV
GitHub Link | Report

UOS

Languages and Databases

Python 3

C

JavaScript

C++

MATLAB

MySQL

Machine Learning

Pandas

NumPy

OpenCV

Scikit-learn

Matplotlib

Plotly

Seaborn

Generative AI

LangChain

OpenAI SDK

Huggingface

LLamaIndex

Deep Learning Frameworks

TensorFlow

Keras

PyTorch

OpenVino

Tools

Git

Docker

Jupyter Notebook

Postman

VSCode

Azure

AWS

Web Engineering

HTML5

CSS3

Bootstrap

FastAPI