Software Engineer · Medical AI Researcher
Ozone Park, New York
Two threads run through my work.
Eight-plus years building production software. Java and Spring Boot, mostly — restaurant management, POS and retail systems, CRM, export/import platforms. Built for the UK, Australian, Bangladeshi and US markets, usually as the lead developer and often as the only one who could see the whole system at once.
Deep learning for clinical imaging. Eight peer-reviewed papers on models that read chest X-rays, brain MRI, mammography, blood smears and clinical photographs. First author on two. The recurring question in all of them is the same one that shows up in enterprise work: what does it take to make this run where it actually has to run?
Currently pursuing a PhD in Information Technology at the University of the Cumberlands, after an MS in Computer Science at Maharishi International University.
Implementations of my published work — each repository holds the architecture, a reproducible training and evaluation harness, and tests.
| Repository | What it does | Venue | Role |
|---|---|---|---|
| ECA-TBNet | Tuberculosis screening from chest X-rays in a 2M-parameter network. Efficient channel attention costs k parameters instead of 2C²/r. |
Discover Artificial Intelligence, 2026 | First author |
| CPE-CA-SWIN | Breast cancer classification across mammography and MRI with one checkpoint. Conditional positional encoding makes the weights resolution-agnostic. | J. Computer Science & Technology Studies, 2024 | First author |
| PharynFuseFormer | Pharyngitis screening from phone throat photographs. CNN and transformer branches fused by bidirectional cross-attention. | Discover Computing, 2026 | Second author |
| QuadCare | One encoder, four screening tasks, per-task calibration, and a plain-language advice layer that refuses to assert anything it isn't sure of. | Frontiers in CS & AI, 2025 | Second author |
| MSAF-BrainNet | Brain tumour MRI classification with per-image multi-scale attention fusion, plus an LLM assistant that drafts a hedged clinical note. | J. Medical and Health Studies, 2023 | Third author |
| EdgeCare | Multimodal diagnosis on CPU-only hospital hardware. 350K parameters, 39 ms median latency, handles missing record fields properly. | Frontiers in CS & AI, 2023 | Co-author |
| iBOT-WBC | Self-supervised pretraining for white blood cell classification, because unlabelled smears are cheap and haematologist time is not. | Discover Artificial Intelligence, 2026 | Co-author |
| HiLT | Oral cancer classification via a patch → region → image hierarchy, with a lesion saliency map that falls out of the forward pass. | Discover Artificial Intelligence, 2026 | Co-author |
Full publication record: ORCID 0009-0009-9124-816X
The models above are mostly smaller than the state of the art, on purpose. A classifier that needs a GPU is not a slightly worse option in a district hospital — it is not an option. Most of the interesting engineering in this work is in what gets removed: attention that costs five parameters instead of sixty thousand, a fusion layer that degrades gracefully when half the patient record is missing, a report generator that stays quiet when the model is uncertain rather than producing fluent prose over a coin flip.
Languages Java · Python · PHP · JavaScript · TypeScript · SQL
Backend Spring Boot · Hibernate · JPA · Node.js · Express · CodeIgniter
Frontend Angular · React · Bootstrap
ML PyTorch · scikit-learn · timm · NumPy · pandas
Data MySQL · PostgreSQL · MongoDB · MSSQL · DynamoDB
Infra AWS · Docker · Kubernetes · Kafka · Jenkins · RabbitMQ
Practice REST/SOAP · microservices · TDD · Agile/Scrum · design patterns
Selected systems built: a CRM replacing HubSpot + DocuSign, including a contactless agreement-signing flow · restaurant management for UK and Bangladesh · retail POS covering stock, finance and supply chain, integrated with e-commerce · doctor's appointment management with pharmacy, prescriptions and SMS · export/import platforms for US and Australian clients · online examination with question banking and MCQ assessment.
PhD, Information Technology — University of the Cumberlands, Kentucky (expected 2029) MS, Computer Science — Maharishi International University, Iowa (2024) MSc, Information Technology — London South Bank University, UK (2012) BSc, Computer Science & Engineering — United International University, Dhaka
Open to collaboration on clinical ML and on backend systems that have to be right.