Projects

Artificial Intelligence

This Project implements a full pipeline to classify and solve equations from images using deep learning and OCR techniques.
It supports both handwritten and typed expressions and reconstructs them into solvable strings.

Proposing a Dynamic Attentive Graph Neural Network for Cardiac MRI Reconstruction in a cascading manner.
Sensitivity map calculation is done according to Compressed Sensing theory using a UNet.
The proposed model successfully reconstructs Cine CMRI and achieves state-of-the-art accuracy with less computation expence compared to previous successfull models.

Designed a Cascading Long Short-Term Memory Proximal Policy Optimization (PPO) model which uses LSTM layers to capture temporal dependencies in stock data and a PPO algorithm to optimize trading decisions.
The environment is from yfinance library with trading data from Jan. 2009 up to Jun. 2024.

A Complete Collection of Deep RL Famous Algorithms implemented in Gymnasium most Popular environments.
Implementation of SARSA and DQN with boltzman in CartPole.
Implementation and comparison of DQN , D3QN and Enhanced D3QN Agents in Lunar Lander environment.
Implemntation of Proximal Policy Optimization algorithm in Swimmer, with clipped objective ppo and adaptive kl ppo agents.

The goal of this project is to develop an enhanced neural network model to classify farsi news articles into their respective categories.
The dataset has been preprocessed with Tokenization and Feature Extraction.

Application of Classic Computer Vision Techniques such as Filtering , Transformation and Feature Extraction for image interpretation.

Designed a controller for a quadcopter to control its flight over boxes in an urban environment, automatically taking precise images of boxes and interpret the images using Computer Vision Deep Learning-Based Approaches.
After the interpretation of the taken image, the quadcopter should determine whether the item matches the target item or not. In case of matching, the quadcopter will land beside the box and turn its front LED’s on.

Implementation of Bug1 , Bug2 and Wall following algorithms for GCTronic’ e-puck from webots.
Each of this algorithms will successfully guide the robot through a maze.
Map of the maze is generated with Bug2 and split-and-merge algorithms.

This Repository includes famous calssification and regression algorithms, each one applied to solve a related problem.
In each problem, Feature Engineering methods are used to prepare raw data for machine learning models by transforming it into relevant features.
The algorithms include K-Nearest Neighbors, Support Vector Machine (SVM), Decision Tree, and Gradient Descent for supervised learning, while Density-Based Spatial Clustering of Applications with Noise (DBSCAN) is used as an unsupervised algorithm.

Controllers for e-puck in Webots enviroment using popular Localization,Planning and Navigation Algorithms.
The Controllers range from simple to more complex ones, providing beginners with better understanding of control process.

The algorithm devised for this game is a combination of a Genetic Algorithm and Minmax, where the Minmax algorithm is used as the fitness function for the Genetic Algorithm.
This game consists of two real-time agents that try to create more walls than their opponent while avoiding collisions with each other and the boundary walls. The Unity framework used in this project is based on Chillin’s monitor games.

Developed a web-based restaurant management system using Django for the backend and React for the frontend.
Implemented features for browsing restaurants, making reservations, and managing user roles with different access levels for customers, restaurant admins, and system admins.

Developed simulations for various types of graphs, including Erdős–Rényi, Watts–Strogatz, and Barabási–Albert models.
Analyzed graph properties such as degree distribution, eigenvalue distribution, and algebraic connectivity, implementing methods for generating and simulating real-world graph behaviors.

A Smart car is built using various sensors, controllers, and an RTOS (Real-Time Operating System) to manage multiple tasks like obstacle detection, cruise control, and light management.
This project was modeled using Tinkercad and simulink and implemented in PlatformIO with FreeRTOS.

Created a version of the classic Doodle Jump game in Assembly 8086, implementing platform generation, player movement, and score tracking.
Developed the game mechanics to ensure smooth and responsive gameplay, with features such as random platform placement and increasing difficulty over time.