Web Development×Agentic AI×Research
I'm Mamun Mahmud — an AI/ML researcher specializing in federated learning and medical imaging, a full-stack developer building scalable web applications, and a competitive programmer passionate about solving challenging algorithmic problems.
About
I'm a software engineer and AI developer who ships full-stack products end-to-end — from database design and backend architecture to polished, performant UIs deployed in production.
I build agentic AI systems with the OpenAI SDK and modern LLM frameworks: tool-using agents, multi-step reasoning workflows, RAG pipelines, and context-aware assistants embedded inside real applications.
On the research side, I work on federated learning, optimization, and medical image analysis as a final-year CSE student at IIUC — ranked 1st in my batch with a 3.91 CGPA. I also write contest code on Codeforces and teach the next cohort of problem solvers.
Education & Experience
- → Ranked 1st in the Batch
- → Top 1% of the Department
- Supported CSE students in algorithmic problem solving and contest preparation.
- Mentored EEE students through C programming fundamentals, debugging, and structured thinking.
- Ran labs and one-on-one sessions to strengthen problem-solving intuition.
- Coordinated bootcamp activities and designed practice problems for contest preparation.
- Helped beginner-to-intermediate learners build confidence in contest-style thinking.
- Trained and reviewed chatbot responses on code- and reasoning-heavy prompts.
- Provided structured feedback to refine output quality and response behavior.
- Organizing team for the Programming Hero NextGen Inter-University Hackathon.
- Team Lead for the 2025 intra-department Tech Hackathon.
- Project Lead and Backend Developer for the IIUC Computer Club website.
Skills
Projects
Resources Management
Full-stack department platform unifying routines, schedules, course materials, classroom reservations, bus info, and faculty contacts for the IIUC CSE community. Built end-to-end with role-based auth, REST APIs, and a relational data model — now augmented with an embedded agentic AI assistant.
IIUC Computer Club Platform
Scalable backend powering club membership, events, and campus activities. Designed the database schema, REST API surface, and deployment pipeline as Project Lead. Type-safe stack with HonoJS + Drizzle ORM on PostgreSQL.
Research & Publications
Adaptive Gradient-Drift-Aware FedProx for Communication-Efficient Federated Learning on Heterogeneous Medical Imaging Data.
- ▸ Federated learning under non-IID data distributions
- ▸ Privacy-preserving medical image analysis
- ▸ Adaptive optimization & communication-efficient training
- ▸ Explainable AI for clinical decision support
Development and Comparative Analysis of Deep Learning Models for Multiclass Classification of Liver Conditions in Ultrasound Imaging
PublishedMamun Mahmud, M. M. Shahriar, M. N. Sakib, F. W. Wibowo
ICSINTESA 2025 — 5th International Conference of Science and Information Technology in Smart Administration, Yogyakarta, Indonesia
pp. 281–286, doi: 10.1109/ICSINTESA68165.2025.11413762
Privacy-Preserving Federated Deep Learning for Multi-Class Liver Fibrosis Staging from Ultrasound Images
AcceptedMamun Mahmud, M. N. S. Rafi, M. R. Uddain, Ts. Dr. G. K. O. Michael, N. Fahad, R. I. Rabbi, E. H. Tusher, F. M. Tarin, Ts. Dr. T. Connie
Camera-ready submitted
Federated Deep Learning for Privacy-Preserving Multi-Class Histopathology Image Analysis
M. M. Shahriar, Mamun Mahmud, M. F. Mridha
Prediction and Detection of Liver Diseases Using Machine Learning and Deep Learning
Mamun Mahmud et al.
HistoSwin-BTNet: A Boundary- and Texture-Guided Swin Transformer Network for Binary Nuclei Segmentation in Histopathology
A. K. M. S. Emtiaz, M. M. G. Hafiz, Mamun Mahmud, M. M. Shahriar, M. K. Morol
AIHA 2026 Workshop @ ICPR 2026 — submitted
Adaptive Gradient-Drift-Aware FedProx for Communication-Efficient Federated Learning on Heterogeneous Medical Imaging Data
Mamun Mahmud, M. F. Mridha
Developing a gradient-drift-aware adaptive FedProx with automatic proximal-coefficient adjustment and client-side regularization. Evaluated on five medical imaging datasets under realistic non-IID settings.
Competitive Programming
Active on Codeforces. ICPC Asia Dhaka Regional contestant ('24 & '25), with a growing record at national and inter-university contests.
Achievements
Let's build something meaningful.
Open to research collaborations, internships, and full-time engineering roles. The fastest way to reach me is over email.