Overview

About

I combine computer vision, Python development, and cross-functional ML rigor to build SegUNet, MedSAM, and dual U-Net pipelines clinicians can explain, while also shipping ML-backed products spanning agritech, GenAI simulations, and security dashboards.

Focus Areas

Research Interests

Responsible Healthcare AI

Designing interpretable SegUNet and MedSAM pipelines that illuminate dental and liver imaging decisions with Grad-CAM, bias checks, and transparent reporting.

Multimodal Diagnostics

Blending panoramic radiographs, CT volumes, agronomic leaf scans, and structured health data to surface clinician-ready, cross-domain insights.

Applied ML Systems

Shipping resilient pipelines for agritech disease detection, cybersecurity automation, and GenAI consulting simulations to prove readiness beyond the lab.

Hands-on

Experience

Aug 2025 – Present · Remote

Neurootix | AI/ML Engineer

  • Designed, tested, and evaluated computer-vision models spanning diverse imaging datasets for production pilots.
  • Ran applied experimentation to steer model selection, pipeline refinement, and research-grade reporting for ongoing AI engagements.
  • Directed Python and image-processing workshops with hands-on coding modules that bridge theory and deployment readiness.

Company Site

Virtual Experience Programs

May 2025 · Remote

British Airways | Data Science Job Simulation

  • Analyzed customer review data to extract themes linked to satisfaction and buying behavior.
  • Built a predictive model to identify the key drivers of purchase decisions.
  • Certificate
  • GitHub
May 2025 · Remote

BCG | GenAI Job Simulation

  • Built a rule-based chatbot prototype to interpret financial data from 10-K and 10-Q reports.
  • Processed structured financial datasets and generated user-friendly GenAI outputs.
  • Certificate
  • GitHub
In Progress

AI Projects

MedSAM FLARE22 Tumor Segmentation

GitHub
Validation Dice ≈ 0.84 · Accuracy ≈ 0.98
  • Fine-tuned MedSAM ViT-B on MICCAI FLARE22 CT scans using MONAI dataloaders and automated slice caching.
  • Generated reusable evaluation artifacts to benchmark focal liver segmentation performance.

PlantVillage Leaf Disease Detection

GitHub
EfficientNet-B0 Test Accuracy 99.25%
  • Hit 99.25% test accuracy with EfficientNet-B0 on 38 PlantVillage crop disease categories.
  • Ensembled EfficientNet-B0, DenseNet121, ResNet18, and ResNet34 with reproducible PyTorch training diagnostics.

Heart Disease Prediction

GitHub
Gradient Boosting · Accuracy 0.8804 · Precision 0.87 · Recall 0.92
  • Built a full ML pipeline covering cleaning, imputation, encoding, feature engineering, and experiment tracking.
  • Benchmarked classical models on clinical risk factors and deployed Gradient Boosting (accuracy 0.8804, precision 0.87, recall 0.92).

Diabetic Retinopathy Dual U-Nets

GitHub
Validation Accuracy 99.9% · DenseNet121 Dice 0.31 vs 0.21
  • Implemented dual U-Net architectures with EfficientNet-B4 and DenseNet121 encoders via segmentation_models_pytorch.
  • Trained with BCEWithLogits + Dice loss, mixed precision, cosine LR schedules, and Adam for IDRiD lesion masks.
  • Preprocessed 384×384 inputs with Albumentations, merged lesion masks, and tracked validation gains (DenseNet121 Dice 0.31).
Product

Web Builds

BookNook

GitHub Live

Responsive e-commerce UI with catalog search, filterable dashboards, and secure checkout flows, designed to keep book discovery, selection, and purchase frictionless.

NightWatch: Link Guard

GitHub Live

Browser-based phishing detection dashboard that scores URLs locally using heuristics, dataset-tuned thresholds, and optional reputation checks inside a cyberpunk SOC-style interface.

NUB Central Social Welfare Club

GitHub Live

Organization hub featuring event galleries and member resources, designed as a responsive portal to strengthen community engagement and accessibility.

Academic

Publication

Enhanced Pediatric Dental Segmentation Using a Custom SegUNet with VGG19 Backbone on Panoramic Radiographs

27th International Conference on Computer and Information Technology (ICCIT 2024) · Jun 2025

Co-authored research on interpretable dental imaging, delivering a custom SegUNet pipeline for pediatric panoramic radiographs.

Read via IEEE
Foundations

Education & Honors

M.Sc Computer Science and Engineering

Brac University

June 2026 – Present

BSc. Computer Science and Engineering

Northern University Bangladesh

Jun 2021 – Nov 2025

CGPA 3.669/4.0

Awards

  • 2nd Place · NUBCC Programming Contest Season 2 (2023)
  • 1st Place · NUBCC Programming Contest Season 1 (2022)
  • Champion · NUB Indoor Badminton (Female) 2022
Toolkit

Skills & Certifications

Skills

  • Programming Languages: Python, Java, C, C++
  • Machine Learning: NumPy, Pandas, Keras, TensorFlow, OpenCV, scikit-learn, PyTorch, Matplotlib, Plotly, Grad-CAM
  • Software Tools: LaTeX, Google Colab, VS Code, CodeBlocks, Logisim, NetBeans, Microsoft Office
  • Database & BI: PostgreSQL, Power BI
  • Core Competencies: Data structures & algorithms, data analysis & preprocessing, image/medical imaging, computer vision, AI agent fundamentals

Certifications

  • Google Cybersecurity Professional Certificate · Credential
  • Biomedical Image Analysis in Python (DataCamp) · Certificate
  • AI Agent Fundamentals (DataCamp) · Certificate
  • Image Processing in Python (DataCamp) · Certificate
Fresh Writing

Latest Blog Posts

I publish deep dives on computer vision, MLOps, and full-stack experiments on Hashnode. Here are a few recent highlights—explore what I am currently building and learning.

Workflow · Kaggle · Colab

Using Kaggle Datasets Directly in Google Colab

A practical playbook for pairing Kaggle API keys with Colab so large competitions stream in directly—no manual zips, just reproducible notebooks and mounted storage.

Read on Hashnode

Analysis · Geopolitics

Code, Chips, and Control

Examining how silicon dependencies, export controls, and software ecosystems turned technology into the primary theater for geopolitical leverage.

Read on Neurootix Blog

Essay · AI Ethics

Data, Power, and the Quiet Conflict

A meditation on the tension between algorithmic decision systems and human agency, and why the next frontier is contested in datasets, privacy laws, and control.

Read on Neurootix Blog

Explore the full archive on Hashnode for long-form case studies and new experiments.

Community

Leadership & Activities

Mar 2024 – Jun 2025 · Dhaka

Vice President | NUB Computer Club

  • Chief Editor of the “Bullet vs Intellect” e-magazine.
  • Successfully organized and led CSE Day 2024 at Northern University Bangladesh.
  • Invigilated the NUBCC Programming Contest 2024 to ensure smooth contest operations.
Mentors

References