Brac University · CGPA: 3.42
Highly motivated CSE student at BRAC University with moderately strong communication and problem-solving skills. I am fluent in Adobe premier pro, illustrator, Photoshop, Google sheets and in coding. Eager to leverage my computer science background and communication skills to excel in any role. I stay up to date with the latest trends and technologies in the field. I also plan to share my knowledge to help others reach their full potential and build successful careers globally.
AHZ associates
Sept 2023 — Sept 2024· 1 yr
Ensuring a better relation between my university and AHZ
Giving interested students proper knowledge about UK higher education
Promoting the brand and help with marketing
Consult with the marketing team to develop long-term commercial plans
Management at educational expos and events
Dot Com MEdia
Aug 2026 — Present· 2 mos
Making sure of proper social media promotion
Coming up with concepts for social media posts
Maintaining page consistency by posting from the FB and Instagram page
2022 – 2026
Brac University · 3.4/4.00
A focused academic workspace for organizing courses, building study notes, capturing spoken ideas, and collaborating with classmates.
Secure registration and sign-in with hashed passwords, HTTP-only sessions, expiry, and protected data access
SQLite persistence with foreign keys, WAL mode, indexes, and seeded demo content
Course creation, course-specific workspaces, colour coding, and safe course removal
Notes with editing, tags, course assignment, favourites, timestamps, search, filters, soft deletion, recovery, and permanent deletion
PDF, image, text, and PowerPoint attachments stored with access-controlled download routes
Browser speech recognition that saves a reviewed transcript as a normal study note
Note sharing with view/edit permissions and collaborator notifications
Persistent one-to-one study-group messaging and an activity centre
Light/dark themes, keyboard focus states, mobile navigation, and responsive layouts
A menu-driven 16-bit x86 assembly project for managing flights, passenger check-ins, seat allocation, flight status, passenger manifests, and final boarding calls.
Features
Displays 15 preloaded flights with destinations and departure times.
Enforces the required step-by-step menu flow.
Stores up to 12 passengers independently for every flight.
Accepts case-insensitive passenger names and stores them in uppercase.
Shows a visual 4-by-3 seat map and prevents double booking.
Supports seat reassignment without corrupting the seat map.
Displays each flight's preloaded status: ON TIME, DELAYED, or BOARDING.
Prints a full alphabetical passenger manifest while keeping seats matched to names.
Validates HHMM time input and triggers a final call from 0 to 10 minutes before departure.
Handles invalid menu choices, flight numbers, names, seats, and times safely.
Campus-focused rental discovery and property management, built for students, teachers, and property owners.
What it does
Searches apartments and parking by area, type, and monthly budget.
Presents price, availability, amenities, commute context, and nearby essentials together.
Supports visit requests and rental applications with an approval trail.
Gives tenants, owners, and administrators role-specific workspaces.
Provides owner listing tools, booking administration, notifications, payments, reviews, and saved commute routes.
Falls back to curated demonstration listings when the API is unavailable, while clearly labelling live availability.
An interactive Rubik’s Cube simulator built with Python, PyOpenGL, and GLUT. Choose a cube size and difficulty, scramble it, rotate faces, track solve statistics, and follow the generated solution.
Highlights
Interactive 2×2×2, 3×3×3, and 4×4×4 cubes.
Easy, Medium, and Hard scramble difficulties.
Animated clockwise, counter-clockwise, and double face turns.
Mouse-controlled 3D camera rotation and wheel zoom.
Automatic scramble generation and inverse-move solution display.
Optional move validation with correction guidance.
Timer, move counter, solve count, and best-time tracking.
Celebration animation after a successful solve.
Adjustable animation speed and built-in help overlay.
A professional university discovery and advising portal built for the CSE370 Database Systems project. Students can search a curated university directory, inspect programmes, create an account, find a
Highlights
Search universities by institution, programme, country, or city
Review programme prerequisites, duration, degree, and fee guidance
Secure student registration using PHP password hashing
Session-based authentication and protected appointment booking
Conflict-safe counsellor scheduling with booking references
Responsive, accessible interface with a shared design system
MySQL schema for XAMPP plus a zero-configuration SQLite demo mode
Technology
PHP 8+
MySQL 8 / MariaDB for the XAMPP workflow
SQLite for quick local demonstration
HTML5, CSS3, and a small amount of vanilla JavaScript
Brain Tumor MRI Segmentation and Classification
A multi-task model that looks at a brain MRI scan and does two jobs at once. It draws the tumor region (segmentation) and it names the scan (classification) as one of four classes: glioma, meningioma, pituitary, or no_tumor. Two architectures are trained and compared, a standard U-Net and an Attention U-Net, both in PyTorch. The dataset is BRISC2025.
The notebook in notebooks/ is the original project, kept with all of its outputs. The src/ folder holds the same pipeline as Python modules.
The problem
Segmentation answers "where is the tumor" by labelling every pixel. Classification answers "what kind of scan is this". A single network handles both: a U-Net encoder and decoder produce the mask, and a small head on the bottleneck features produces the class. The two losses are combined as 0.7 times Dice for the mask plus 0.3 times cross-entropy for the class.
Data
BRISC2025 ships a segmentation task and a classification task. The segmentation task holds tumor scans paired with masks. The classification task holds scans in four class folders, including healthy scans under no_tumor. Images are grayscale and are resized to 256 by 256, denoised with a light Gaussian blur, and the masks are cleaned with small morphological open and close steps.