misc

My research journey, thoughts, and miscellaneous insights.

My Research Journey

The Beginning: Epidemiological Modeling (2023)

My research journey began in my second year at BUET with Prof. Dr. K.M. Ariful Kabir, where I explored epidemiological modeling. We developed stochastic, agent-based models on multiplex social networks to simulate disease spread in human populations. This work opened my eyes to how computational methods could address real-world crises beyond theoretical frameworks.

Key Learning: The power of simulation in understanding complex socioeconomic effects on public health.

Publication: Study on Locomotive Epidemic Dynamics in a Stochastic Spatio-Temporal Simulation Model on a Multiplex Network — arXiv, 2025

Transition to Medical AI: Semi-Supervised Learning (2023–2024)

Working with Prof. Dr. M. Sohel Rahman, I transitioned into medical AI through semi-supervised and multimodal breast cancer classification. This project combined imaging data with clinical and demographic information, fundamentally shaping my understanding of fairness, data bias, and interpretability in healthcare AI.

Key Learning: The critical importance of building ML systems that are actually safe for human health data.

Publication: Breast Cancer Detection Using Semi-Supervised Learning: Multimodal Data Integration and Comparative Demographic Analysis — NSyS 2024

In parallel, I worked with Prof. Dr. M. Saifur Rahman on Fair-RAG-IM, a bias-aware retrieval-augmented generation framework for clinical decision support, focusing on interpretability and equitable medical decision-making.

International Collaborations: Cryo-Electron Tomography (2024–Present)

Carnegie Mellon University — Xu Lab

In mid-2024, I began collaborating remotely with the Xu Lab at Carnegie Mellon University. My work focuses on unsupervised Cryo-ET segmentation and macromolecule localization in extremely crowded and noisy cellular tomograms.

Research Activities:

  • Denoising and preprocessing 3D tomograms
  • Particle picking under minimal labeling
  • Building training-free or weakly supervised segmentation pipelines
  • High-resolution biological structure reconstruction

Key Publications:

More recently, I started collaborating with Dr. Muyuan Chen at Stanford University on PFIB tomography of biological samples, particularly Hydra tissue. This work involves segmentation and structural analysis of high-resolution PFIB volumes to interpret morphology and organization from 3D datasets.

Key Learning: The same underlying challenge exists across different scales — recovering meaningful biological structure from extremely challenging imaging data.

Contextualized Machine Learning (2024–Present)

Since late 2024, I’ve been working with Dr. Ben Lengerich at the University of Wisconsin–Madison on a comprehensive review of contextualized machine learning. The core insight is that many models assume one global rule for everyone, but real-life decisions depend on context: patient history, environment, measurement conditions, and more.

I contribute to AdaptInfer, which aims to build models that can:

  • Adapt to shifting contexts
  • Explain behavior instead of acting like black boxes
  • Provide interpretable and contextual decision-making

Publication: Contextualized Machine Learning: Towards Adaptive, Interpretable, and Generalizable AI Systems — UW–Madison Research Project, 2025

Specialized Applications: BCI and Real-Time Systems (2024–2025)

EEG-Based Epilepsy Detection

With Prof. Dr. A.B.M. Alim Al Islam, I developed NEUROSKY-EPI, the first open single-electrode epilepsy EEG dataset with context-aware modeling for real-time brain–computer interface applications.

Publication: NEUROSKY-EPI: The First Open Single-Electrode Epilepsy EEG Dataset with Context-Aware Modeling — NeurIPS 2025 Workshop on Time Series for Health

Real-Time Drone Detection

Working with Prof. Dr. Ch. Md. Rakin Haider, I contributed to SpectraSentinel, a dual-stream, lightweight real-time drone detection and tracking system combining spectral and spatial fusion cues.

Publication: SpectraSentinel: Lightweight Dual-Stream Real-Time Drone Detection, Tracking and Payload Identification — arXiv, 2025

Current Focus: Undergraduate Thesis (2025–Present)

My undergraduate thesis is jointly supervised by Prof. Dr. M. Sohel Rahman and Prof. Dr. Md. Shamsuzzoha Bayzid. I’m working on:

  • Low supervision and training-free segmentation pipelines for cryo-electron microscopy data and connectomics
  • Multimodal RNA segmentation and task accuracy analysis for biological connectomics
  • Goal: Extract useful biological understanding from messy imaging and sequence data with minimal manual labeling

Entrepreneurship: From Research to Impact (2025–Present)

PinkLifeLine — Health-Tech Startup

Co-founded PinkLifeLine as a health-tech startup emerging from our NeoScreenix project, which won the Global Champion award at the Johns Hopkins Healthcare Design Competition 2025.

Key Achievements:

  • Funded by Bangladesh National ICT Division
  • Partnerships with Sustainlaunch Labs (global innovation accelerator) and Herwill (women’s empowerment organization)
  • Leading a 50+ medical professionals network
  • Developing ML clinical data pipelines and screening workflows

Media Coverage:

Research Philosophy

My work has crystallized around four core principles:

1. Structure from Chaos

How do we extract meaningful structure from extremely difficult biological imaging data? Whether it’s cryo-ET tomograms, PFIB volumes, or connectomics data, the challenge remains: making sense of noisy, complex, high-dimensional biological information.

2. Context Matters

Machine learning systems need to be contextual, fair, and clinically useful. A model that works for one population or setting may fail catastrophically in another. Context-aware AI is not just better science—it’s an ethical imperative.

3. Minimal Supervision, Maximum Impact

The most interesting biological questions often come with minimal labeled data. How do we build systems that can learn from sparse supervision while maintaining reliability for critical applications?

4. From Bench to Bedside

Research should ultimately serve human health and scientific understanding. Whether through startup ventures like PinkLifeLine or open datasets like NEUROSKY-EPI, I believe in translating research into real-world impact.

Looking Forward

I continue to collaborate with groups at Carnegie Mellon University, University of Wisconsin–Madison, Stanford University, and BUET. My long-term goal is to build machine learning systems that can both interpret complex biological data and support real decision-making in health and science, while being aware of bias, context, and uncertainty.

Current Active Projects:

  • Cryo-ET segmentation with foundation models (CMU)
  • PFIB tomography analysis (Stanford)
  • Contextualized ML review and AdaptInfer framework (UW-Madison)
  • RNA segmentation and connectomics (BUET thesis)
  • PinkLifeLine clinical deployment

“The most beautiful thing we can experience is the mysterious. It is the source of all true art and science.”

— Albert Einstein