Md Mofazzal Hosen · Software Engineer
I build full-stack and applied-ML systems for business operations, incident response, medical imaging, and hard perception problems. My best work makes probabilistic AI useful without giving it unchecked authority.
Selected projects · 5 of 18
A recruiter-friendly shortlist. Each project shows the problem, my ownership, technical choices, evidence, and honest limitations.
An autonomous incident-response system where AI investigates and proposes, while deterministic Java guardrails retain authority.
Java 25 · Spring Boot · Next.js · PostgreSQL · +4
A controlled ablation showing which architecture changes actually help find people smaller than eight pixels in disaster imagery.
PyTorch · YOLO11 · CBAM · SAHI · +2
A FastAPI service for Bengali/English transcription and structured lab-report extraction with deterministic offline replay.
Python · FastAPI · Pydantic · OpenAI · +2
A two-sided Android pharmacy app with customer checkout, a live admin catalog, design patterns, automated tests, and CI builds.
Java · Android · Firebase · JUnit · +2
An OMNeT++ simulation of pooled GPU leasing across two campus network segments, with capacity beacons, centralized scheduling, background traffic, and repeatable analysis.
C++ · OMNeT++ · Python · pandas · +2
Browse all 18 projects across production systems, applied AI, research, web, mobile, hardware, and coursework.
Experience
Roles, scope, and concrete outcomes—with proprietary details kept appropriately bounded.
MAEKNIT Inc. · New York, USA · Remote
Building Python-based Odoo ERP modules on PostgreSQL for live business-operations workflows.
Algosoft Technologies Limited · Dhaka, Bangladesh
Shipped a multi-module enterprise ERP/DMS spanning export/import compliance, reporting, support, and device tracking.
Research
Submitted manuscripts, active studies, and completed academic work are labeled separately so evidence is easy to audit.
Range-Covering Deep Learning addresses scarce high-heart-rate training coverage by adapting a 76K-parameter wrist-PPG estimator with distribution and spectral targets. Adding 59 minutes of treadmill recordings reduced held-out MAE from 18.0 to 11.9 bpm and reduced error by 19%–40% across five external datasets, while held-out PPG-DaLiA subject error did not change detectably.
Submitted · under review
Education & leadership
Khulna University of Engineering & Technology (KUET)
Education and academic workTechnical Lead
General Secretary
Junior Executive
Contact
If you are hiring for full-stack, backend, applied-AI, or research engineering, send the role and the problem you need solved. I normally reply within one business day.
Best way to reach me
mofazzalhossen874@gmail.com