Meet Fahad Rafique: A
Creative Portfolio

As a devoted learner and aspiring AI student, I'm motivated by a deep interest in deep learning, with a specific concentration on the fascinating field of computer vision. My knowledge encompasses reinforcement learning and deep learning in addition to computer vision, which I use to solve challenging data problems and create game-changing models. I enjoy the chance to work with others on complex issues while putting my skills to use to significantly advance our common goals. Beyond AI, I also thrive at using Power BI to leverage data insights, turning raw data into dynamic dashboards that encourage reasoned decision-making.
The future of AI and data-driven insights will be shaped by innovation and creativity, so come along with me on this trip.

Enhancing Blackjack Strategy through Reinforcement Learning

The project implements traditional blackjack tactics with a focus on card counting and the basic strategy. Monte Carlo techniques is used to further improve blackjack strategy. The experimental results provided insightful information about how various parameters affect the performance of the agents. Both on-policy and off-policy techniques yielded comparable percentages of wins for the fundamental strategy, but the on-policy strategy consistently generated larger average winning bankrolls.

Ball Trajectory Simulation using Kalman and Particle Filter

This project revolves around the advanced prediction of ball trajectories using the Kalman and Particle filters. The primary objective was to leverage these filters—both linear and non-linear—in order to understand their impact on ball trajectory forecasting. The project delved into the fascinating realm of how these filtering algorithms could be adapted to account for different launch conditions, observation intervals, and instances of observation failures. By exploring these nuances, the project shed light on the adaptability and robustness of the filters in real-world situations.

Crop Load Estimation using Object Detection

Estimating the crop load is essential for maximizing orange crop productivity. I developed a system using computer vision methods and deep learning models that could count oranges with an accuracy of 94% in 0.2 seconds. Orange dataset is collected manually from Punjab Pakistan and annoated using labelImg. Faster RCNN model is used and TFOD api is used for inferencing.

Skimlit "Abstract Classification"

In this project, NLP model to classify abstract sentences into the role they play (e.g. objective, methods, results, etc) to enable researchers to skim through the literature and dive deeper when necessary. Hybrid models and transfer learning is used for training.

Human Step Detection
(Step Made or Not) using Pytorch

This project uses accelerometer and gyroscope data to determine whether the step was made or not. This project's major goal is to overcome the unbalanced dataset. To solve this issue, a weighted loss function with cross-entropy on unbalanced data is used. Finally, the model had an accuracy of 88% when making predictions.

Classification
using TensorFlow

With the help of TensorFlow, a simple CNN model that can predict whether an image is of a dog or a cat is built from scratch. A model has a 95% prediction accuracy rate, which is excellent intuition for classifying problems.