About
Results-driven AI Automation Engineer & Full-Stack Architect specialising in designing production-ready LLM ecosystems, scalable RAG pipelines, and enterprise SaaS architectures. Currently building AI software solutions for government-level projects at MySoftHaven (BD) Ltd., and serving as a remote Research Assistant at the Centre for UntieAI Artificial Intelligence Research Lab, CU Boulder.
My research focuses on multimodal AI, medical imaging (radiology benchmarks), vision-language models, and NLP for under-resourced languages. I have co-authored 10+ papers spanning KDD, ACL, MICCAI, SaTC, ICMLAS, Health, and Plos One.
I also teach AI/ML laboratory sessions at North South University, where I completed my BSc in CSE with a 3.81 GPA (Summa Cum Laude).
Skills
AI / ML Frameworks
Vision-Language & Multimodal
LLMOps & RAG
Backend & Cloud
Frontend
Languages
News
2025 — "From Benchmarks to Deployment: A Calibrated, SHAP-Guided, Production-Ready Pipeline for Diabetes Risk Prediction" accepted at ICCIT 2025 (First Author). Accepted
2026 — "RA Biomarkers: Consensus Feature Selection Pipeline" accepted at ICMLAS 2026 (First Author). Accepted
2026 — Two papers accepted at SaTC 2026: "Engineering Spatiotemporal Forecasts at Scale: Dhaka Weather…" and "Harnessing Vision Language Models for Negation Sensitivity in VQA". Accepted
2026 — "A Leakage-Safe Multimodal Medical Generalist Benchmark: Chest X-Ray Classification…" submitted to MICCAI 2026. Submitted
2026 — "The Differential Sensitivity Fusion Approach to Layerwise DCNN Pruning" submitted to Plos One. Submitted
2026 — "Decentralizing Reputation Scoring Using Blockchain Behavioral Data and ML" submitted to MMAR-2026. Submitted
2025 — "RadMultiBench: A Multimodal Benchmark Revealing the Limits of Frozen VLMs in Radiology" under review at Health (journal) (First Author). Under Review
2025 — "Bridging the Gap in Low-Resource Languages: Multilingual Multimodal Hate Detection in Memes" under review at Plos One (First Author). Under Review
2025 — "Face2Profile: A Face-and-URL Dataset for Open-Ended Profile Construction" under review at KDD 2026 Datasets & Benchmark Track. Under Review
2025 — "Color-Aware Swin Transformers for Tea Leaf Disease and Pest Classification" under review at Artificial Intelligence in Agriculture. Under Review
Publications
10+ papers2025
2026
2026
2026
MULTI
BENCH
2026
One
2026
One
2026
2026
Agriculture
Live Apps
Experience
- Develop AI software solutions for government-level projects — end-to-end LLM integration, RAG pipelines, and workflow automation.
- Architect production AI ecosystems using FastAPI, FAISS, Gemini 1.5 Flash, and Docker, reducing manual data processing time by 60%+.
- Design multi-tenant SaaS architectures with automated CRM sync concepts for Salesforce and HubSpot.
- Constructed automated data pipelines to process 500 GB+ of complex multimodal medical imaging and NLP datasets, cutting data preparation time by 50%.
- Co-authored 3 high-impact papers currently under review at KDD, ACL, and SaTC by driving experimental design and scalable AI methodologies.
- Benchmarked state-of-the-art VLMs (CLIP, ViLT, BioViL-T) across 10,000+ data points to identify architectural bottlenecks in frozen radiology models.
- Developed automated evaluation scripts for RadMultiBench, accelerating multimodal medical model testing by 3×.
- Architected scalable multi-tenant cloud infrastructure on AWS and GCP using Docker and Kubernetes, achieving 99.9% system availability.
- Accelerated deployment cycles by 40% via automated CI/CD pipelines for enterprise AI applications.
- Optimised REST APIs and PostgreSQL databases, cutting query response time by 35% and enabling 5,000+ concurrent requests.
Teaching
- Instruct core AI/ML lab sessions for 50+ students per semester, delivering hands-on PyTorch and TensorFlow demonstrations — average class performance up 15%.
- Engineered an automated Python-based grading system, reducing evaluation time by 80% while providing instant, accurate feedback.
- Teaching scope: Data Structures & Algorithms, OOP, C++, Java.