NLPTransformersDeep Learning
Understanding Transformer Architectures in Modern NLP
A deep dive into the attention mechanism and how transformers revolutionized natural language processing, from BERT to GPT and beyond.
March 15, 20248 min read
Shohanur Islam Sobuj
Hi, I'm
Shohanur Islam Sobuj.
I build _.
I'm a machine learning engineer specializing in AI, NLP, and computer vision. Currently, I'm focused on developing scalable ML infrastructure and conducting research in parameter-efficient fine-tuning of large language models.
01.
About Me
Hello! I'm Shohanur, and I enjoy creating intelligent systems that solve real-world problems. My journey into AI began with curiosity about how machines could learn and has evolved into expertise in Natural Language Processing, Computer Vision, and Biomedical Image Processing.
Fast-forward to today, and I've had the privilege of working at companies ranging from early-stage startups to established enterprises. My work spans from academic research published in top-tier venues to production-ready systems deployed at scale, handling millions of requests daily.
When I'm not at the computer, I'm usually exploring the latest developments in AI, contributing to open-source projects, or sharing knowledge through technical writing and mentoring aspiring ML engineers.
Here are a few technologies I've been working with recently:
Python
PyTorch
TensorFlow
Transformers
Docker
Kubernetes
Apache Kafka
MLOps
AWS/GCP
02.
Experience
Anymate Me GmbH
Current
Machine Learning Engineer
Dec 2024 - Present (10 m)Köln, GermanyFull-Time
Developing and deploying MLOps-driven ML solutions to streamline business processes
PythonTensorFlowPyTorchMLOpsDocker
Business Automation Ltd.
Machine Learning Engineer
Nov 2023 - Oct 2024 (1 yr)Dhaka, BangladeshFull-Time
Built scalable ML infrastructure and microservices. Managed CDC with MySQL using Debezium, Kafka, and Zookeeper. Delivered SmartRemarks and OCR-based TIN certificate validation system.
PythonKafkaDockerMySQLOCRNLP
Anchorblock Technology LLC
Machine Learning Engineer
May 2022 - Oct 2023 (1 yr, 6 m)Dhaka, BangladeshFull-Time
Developed large-scale distributed machine learning systems. Implemented MLOps principles including CI/CD for ML. Built diverse AI bots and designed APIs for ML/AI model integration.
PythonMLOpsCI/CDAPI DesignDistributed Systems
03.
Featured Projects
L-TUNING: Synchronized Label Tuning
Published at ICLR 2024
Novel tuning methodology
Performance improvements on NLP benchmarks
A novel approach for prompt and prefix tuning in Large Language Models with synchronized label optimization. Published at ICLR 2024 Tiny Papers.
PyTorchTransformersLLMsFine-tuning
Rice Leaf Disease Classification
Transfer learning approach
High classification accuracy
Agricultural AI application
Leveraging Pre-trained CNNs for efficient feature extraction in rice leaf disease classification achieving high accuracy on agricultural datasets.
TensorFlowComputer VisionCNNTransfer Learning
Contrastive Learning for NLI
Zero-shot performance
Cross-lingual capabilities
EMNLP 2023 publication
Universal Zero-Shot Natural Language Inference with Cross-Lingual Sentence Embeddings using contrastive learning. Published at EMNLP 2023.
PythonNLPContrastive LearningMultilingual AI
04.
Publications
05.
Latest Writing
Featured Posts
MLOpsDevOpsProduction ML
MLOps Best Practices for Production ML Systems
Lessons learned from deploying machine learning models at scale, including CI/CD pipelines, monitoring, and infrastructure considerations.
February 28, 202412 min read
Recent Posts
Computer VisionMedical AIHealthcare
Computer Vision for Medical Imaging: OCT Analysis
Exploring how deep learning techniques can be applied to optical coherence tomography images for diabetic retinopathy diagnosis.
February 10, 202415 min read
Zero-Shot LearningNLITransfer Learning
Zero-Shot Learning in Natural Language Inference
How contrastive learning enables models to perform NLI tasks without task-specific training data, with practical implementations.
January 22, 202410 min read
InfrastructureKafkaDocker
Building Scalable ML Infrastructure with Kafka and Docker
A practical guide to setting up event-driven ML pipelines using Apache Kafka, Docker containers, and microservices architecture.
January 8, 202414 min read
LLMsFine-tuningTransfer Learning
Fine-tuning Large Language Models: A Comprehensive Guide
From prompt tuning to parameter-efficient methods, explore different approaches to adapt LLMs for specific tasks and domains.
December 18, 202318 min read
06.
Get In Touch
I'm always open to collaboration and new opportunities in AI research and development. Feel free to reach out if you'd like to discuss machine learning, research, or potential collaborations.
Or reach out directly at shohanur.ai@gmail.com
Prefer other ways to connect?