Generative AI, Foundation Models, LLM Fine Tuning, Model Development, Insight generation and Model Infrastructure for Consumer Credit Risk Management of 100M customers to influence budget planning and federal compliance.

Developing efficient, effective, and robust adversarial attacks and defenses for Deep Neural Networks, Deep Reinforcement Learning (DRL), Transfer Learning, and Federated Learning.Worked with industry leaders: Dr. Ananthram Swami (Presidential Award Winner Researcher) and Dr. Hoda Eldardiry (Associate Professor VT)https://scholar.google.com/citations?user=3dfMSRMAAAAJ&hl=enhttps://scholar.google.com.pk/citations?hl=en&user=wToVkEUAAAAJ

I am assisting Dr. Lifu Huang (UIUC Doctorate, https://wilburone.github.io/) for the course Machine Learning (CS4824/EE4424)Topics:Overview of machine learning: learning from data; overfitting, regularization, cross-validation.Supervised learning: decision trees, logistic regression, linear regression, naive Bayes, support vector machines, neural networksUnsupervised and semi-supervised learning: clustering; principal components analysisDeep learning: convolutional neural networks, recurrent neural networksReinforcement learning, machine learning applications (natural language processing, recommendation)

• Forecasting the annual banking loss in the risk management team and explaining the causes of loss using causal inference and explainable AI techniques in Spark, AWS, DataBricks, Python and DoWhy.• Impact: More informed business decisions are being taken after knowing the causes of Capital One’s annual loss of 8 billion dollars, preventing the loss and increasing the revenue.• Achievement: Deployed these techniques on a Machine Learning service (MLaaS) platform to help other teams.

• As a data scientist, I have worked on developing an E-Payment Fraud Engine, to predict real-time fraudulent transactions. We used multiple transactional, merchant, and region-based features. We also used cardholder historical profiling stats as well as time-based features to predict fraudulent activities. We were primarily using Python, PySpark, Spark, Hadoop, Scala, and Java languages for the project.• Have experience in data preprocessing and feature engineering. Skilled in working with machine learning models namely decision trees, random forests, gradient boosted trees, and logistic regression.• Have experience of working on Jupyter Notebook and several libraries some of which are Pyplot, Pandas, Numpy, Scikit and MLlib.

I worked on mainly two data analytics, networking, and systems-based projects in Networks, Systems and Security Group (NSG) with results best for publication:1) Performance characterization of video streaming on low-end phones Supervisors: Dr. Zartash Afzal Uzmi (Stanford Doctorate, Confiz Director, Pak Wheels Director), Dr. Ihsan Ayyub Qazi (Pittsburgh Doctorate), and Dr. Zafar Ayyub Qazi (UC Berkley PostDoc). One measurement study was on Netflix and the other was on Youtube with data collection and data analytics in Python. Results based on data analytics suggested low-end phone level optimizations to both sites.Keywords: Adb Shell, Selenium, Matplotlib, Keras, Pandas, Youtube, Netflix, Video streaming, Automation, Measurement study Conference targeted: ACM SIGCOMM 20202) Reduced end-user latency for the Software-defined enterprise data centers using Machine LearningSupervisors: Dr. Zartash Afzal Uzmi (Stanford Doctorate, Confiz Director, Pak Wheels Director), and Dr. Muhammad Fareed Zaffar (Duke Doctorate, Machine Learning Pioneer)Keywords: Proactive flow rule installation, Flow popularity approach, SVMs, Imbalanced Data, Pipeline designing, scalability in a real environment, ground truth and preprocessing, Big Data, Ryu Controller, Mininet, real-time classification, LSTMs, RNN, GRU, Keras, Python, SMOTE, oversampling, Pandas.Conference targeted: INFOCOM 2020Note: Publication material can be shared on request

- Computer Vision and Graphics Lab (CVGLab)I worked on 'Scene Classification for autonomous systems' on our own unique collected dataset on the roads of Pakistan and Canada under the supervision of Assistant Professor Dr. Murtaza Taj. We have collected data based on six classes or places of the developing country Pakistan and developed country Canada. After preprocessing, models of Keras like ResNet and VGG were used for training. 93% accuracy was achieved.

Probability Math 230 - Associate Professor Dr. Shaheena Bashir - https://lums.edu.pk/lums_employee/6164 I am working for probability with Associate Professor Dr. Shaheena and highlighting the importance of probability which has applications in every field such as Data Science by giving tutorials/seminars and making applied questions/assignments.

Discrete Mathematics CS 210 - Assistant Professor Dr. ImdadUllah Khan - Rutgers, The State University of New Jersey - https://lums.edu.pk/lums_employee/4100Due to my exceptional academic performance in Discrete mathematics, I worked with Assistant Professor Dr. Imdadullah Khan and introduced new topics such as cryptography and hashing in order to meet the requirements of the ever-changing field of computer science.

Lahore School of AI is a community of 1200+ AI practitioners spread across the globe working in tandem to solve real-world problems. It is a part of the global community "School of AI" and their mission is to offer a world-class AI education to anyone on Earth for free. We organized the 1st Artificial Intelligence Meet-up event of Lahore AI on 5th April, 2019, and invitations were opened for every AI enthusiast from industry and academia with speakers from IBM, FAST university & Google Developer Expert from Nepal. (www.facebook.com/groups/LahoreAICommunity/ | www.folio3.ai/events/)

Worked for better AI solutions in AI research group and detailed improvements in Afiniti product.Keywords: R, Python, Pandas, Call center Environment, .NET, Bayesian models, feature engineering. AT&T Mexico client.

DialedIn is a smartphone application made in Silicon Valley USA and it is an advanced form of Powerpoint that is used in any conference to improve the communication among the audience and between the audience and presenter. This is currently running in the USA and is being advertised at international conferences in Pakistan by us.

E-Dukaan is an online platform aiming to serve universities students with high quality mobile accessories at lower rates than the market price. I am currently running E-Dukaan in LUMS, ITU, UMT and FAST. My aim is to take this startup to new heights by increasing the variety of products, marketing them and by gaining our customer’s trust. My Leadership and motivational skills have given my startup a great head start. A team of 20 people is working for me. I am experienced to encourage and motivate my staff to maximize the work and sales. https://www.facebook.com/EDukaan.org/

Models of Automation of tax collection, tax monitoring and data analytics solutions were proposed and some were implemented to minimize effort and maximize government’s revenue. These include basic flaws in the system, drones monitoring and blockchain applications.