Kazi Ababil Azam

Towards Intelligent Traffic Signaling in Dhaka City Based on Vehicle Detection and Congestion Optimization

Supervisors: Dr. A. B. M. Alim Al Islam

This was my undergraduate thesis project under the supervision of Dr. A. B. M. Alim Al Islam, where we proposed a system architecture for an intelligent traffic signaling system suited for the non-lane based, heterogeneous traffic in Dhaka city. The single-intersection traffic signaling system uses object detection and multi-objective optimization on a Raspberry Pi. The devised system architecture was developed into a prototype, which was then tested at a five-road intersection in Dhaka, Bangladesh. We have worked on real-time video processing, object detection in low-resource environments, and multi-objective optimization algorithms. My personal contributions to the project include the dataset preparation to train the model, the evaluation and selection of the object detection model, the overall integration of the cyber-physical system, and the testing and evaluation of the system at the intersection. [Manuscript on arXiv]

Investigating Data Privacy in AI Chatbots

Supervisors: Dr. Dipto Das, Dr. Imtiaz Karim

Sensitive AI chatbots are becoming increasingly popular, but they raise significant concerns based on usage patterns. In this project, we aim to investigate the data privacy of these chatbots by analyzing posts about using these chatbots on the internet. We are collecting anonymous posts related to romantic chatbots and AI partners from Reddit, which we are analyzing to understand the shared data and concerns regarding the privacy of these chatbots.

Pedestrian Trajectory Categorization and Analysis in AV Simulation

Supervisors: Dr. Golam Md. Muktadir, Dr. A. B. M. Alim Al Islam, Dr. Mahmuda Naznin

In simulation-based testing of autonomous vehicles, it is essential that simulation models generate diverse pedestrian trajectories and cover a wide range of real-life possible situations. As a result, there is a need for metrics to assess whether the synthetic trajectories produced by simulation models truly represent all kinds of real-world trajectories. To address this gap in literature, we propose formulating such metrics by analyzing real-world pedestrian trajectory datasets. Currently, we are categorizing the trajectories through finding structures based on different algorithms, including human annotation over the trajectories.

Evaluating Reasoning in Large Language Models

Supervisors: Dr. Golam Md Muktadir

Large language models (LLMs) have shown impressive performance in various tasks, the newer versions with reasoning capabilities are even more impressive. But are these models really capable of reasoning through a complicated problem and finding the solution? In this project, we are challenging reasoning and regular LLMs to solve different complex problems which require logical reasoning and perceptional skills.

NHT1071: A Traffic Image Dataset for Non lane-based and Heterogeneous Traffic for Adaptive Traffic Signal Scheduling

Supervisors: Masfiqur Rahaman, Dr. A. B. M. Alim Al Islam

The Non-lane based Heterogeneous Traffic dataset NHT1071 was prepared with 1071 traffic images from different locations in Dhaka city, taken in various times of the day. The dataset was prepared to train the object detection model for vision-based traffic signaling systems. We labeled the dataset with the help of human annotators and trained object detection models with the help of the labeled dataset, the best of which was used in the traffic signaling system project above.

SyncInc

A web-based project and task delegation management software designed for organizations. Stack includes PostgreSQL as the database, Django for the backend, ReactJS with Material UI for the frontend, Django REST framework for API development, and Django Channels for real-time notifications.

Contradictory, My Dear Watson (Extended)

A project extended from a Kaggle competition to detect contradictions in multilingual text data, including Bangla text, through rigorous preprocessing and fine-tuning.

Laser Security System With Arduino

A project to control a laser security system with microcontrollers, where the ATMega32 unit manages the laser and the Arduino unit controls the alarm system and overall coordination.

Ray Tracing and Illumination Using OpenGL

Implementation of an image generation pipeline using ray tracing and illumination techniques to create realistic images of geometric shapes (sphere, pyramid, cube) and a 2D plane using OpenGL.

Asteroids

5th position (out of 15 teams in the final round, 68 teams overall) by developing 'Asteroids', a level-based top-down shooter game using p5.js.

Implementation of a Congestion Control Algorithm for TCP Flows in ns-2

Implementation of a research paper on congestion control in ns-2 codebase, specifically in TCP to achieve improved throughput in wired and wireless networks.

Compiler

Built a compiler from scratch including steps for creating a symbol table, building a lexical analyzer, semantic analyzer, and generating intermediate code.

Organizer, BUET CSE Fest 2023

Jun 2023 - Jul 2023

A proud part of the organizing team of BUET CSE Fest 2023, the 2023 edition of the annual departmental festival consisting inter-university competitions such as hackathon, programming contest, deep learning competition, CTF competition and extra-curricular events such as cultural program, dance mob, treasure hunt, indoor and outdoor games.

Director of Events, BUET Cyber Security Club

Jun 2023 - Apr 2024

Served as the Director (Events) of BUET Cyber Security Club, a student-run organization under the Department of CSE, BUET. I was responsible for promotion and organization of cybersecurity related events, seminars and both intra and inter-university Capture The Flag (CTF) competitions multiple times through-out the tenure.