👋 Hey there, this is...

Nahian Ibn Asad

|

Fascinated by turning questions into discoveries, I explore the frontiers of Machine Learning, Deep Learning, Computer Vision, and Natural Language Processing to shape the next generation of Artificial Intelligence.

About Me

AI enthusiast turning curiosity into research and innovation

With 5 years of experience in Machine Learning, I am fascinated by the evolving landscape of Artificial Intelligence and its potential to transform the way systems perceive, understand, and learn. My work spans Machine Learning, Deep Learning, Computer Vision, Natural Language Processing, Large Language Models, Vision Language Models, Vision Transformer where I continuously explore new ideas and approaches to solve challenging problems.

Driven by curiosity and a passion for discovery, I focus on building robust models, designing meaningful experiments, and deepening my understanding of intelligent systems. I aim to contribute to research that bridges theory and real-world impact.

Machine Learning

Deep Learning

Artificial Intelligence

Computer Vision

Natural Language Processing

LLM

VLM

Few-shot Learning

CLIP

Transformer

IGNN

Transformer

Years Professional Experience

Awards

Education

Bachelor of Science in Computer Science and Engineering

Islamic University of Technology (IUT) • 2019-2023

CGPA: 3.92 / 4.00

Position: 9th

Thesis

Recognizing Traffic Signs using Fine-tuning Based Few-Shot Object Detection

Languages

Bengali

Native

English

Fluent

Research Work

Exploring the frontiers of technology through innovative research and publications

Achieved up to 2.4× mAP improvement over state-of-the-art FSOD models on BDTSD dataset

FUSED‐Net: Enhancing Few‐Shot Traffic Sign Detection with Unfrozen Parameters, Pseudo‐Support Sets, Embedding Normalization, and Domain Adaptation

Developed FUSED-Net, a domain-adaptive traffic sign detection model optimized for few-shot learning with limited data.

Traffic Sign RecognitionFew-Shot LearningObject DetectionDomain AdaptationFaster R-CNNMachine LearningComputer VisionEmbedding NormalizationModel GeneralizationDeep LearningCross-Domain LearningFew-Shot Object Detection (FSOD)

Learn More

Highlighted key weaknesses in QA models under adversarial perturbations

Contextual Breach: Assessing the Robustness of Transformer-based QA Models

Created an adversarially-augmented SQuAD dataset to benchmark QA model robustness against diverse perturbations.

Contextual Question-AnsweringAdversarial PerturbationsRobustnessAdversarial NoiseSQuAD DatasetTransformer ModelsRobustness MetricsNoise Intensity LevelsQuestion-Answering ModelsDeep LearningModel Vulnerabilities

Learn More

Improved segmentation accuracy by modeling complex spatial dependencies in images

ImpliSeg: Implicit Graph Neural Networks for Semantic Segmentation

In Progress

Developed an Implicit Graph Neural Network (IGNN) to enhance semantic segmentation by capturing long-range dependencies and contextual relationships within images.

Implicit Graph Neural NetworkSemantic SegmentationContextual RelationshipsGraph-based LearningDeep LearningComputer VisionLong-range DependenciesNode EmbeddingsImage UnderstandingNeural Networks

Enabled accurate modeling of complex object interactions and relationships within visual scenes

GraphVQ: Scene Graph Generation with VQ-VAE

In Progress

Implemented a Vector Quantized Variational Autoencoder (VQ-VAE) to generate structured scene graphs from images, capturing object relationships and contextual semantics for enhanced image understanding.

VQ-VAEScene Graph GenerationImage UnderstandingObject RelationshipsContextual SemanticsDeep LearningGraph RepresentationsComputer VisionStructured RepresentationGenerative Models

Demonstrated reduced computational load and faster inference while preserving high-quality LM responses

PromptShrink: Efficient Web Agent via Minimal LM Prompts

In Progress

Developed a web agent that interacts with large language models using highly optimized, minimal-length prompts, reducing computational cost while maintaining response accuracy and relevance.

Web AgentPrompt OptimizationLarge Language ModelsEfficient AINatural Language ProcessingDeep LearningConversational AIPrompt EngineeringResource-efficient AILM Optimization

My Projects

Here are some of my projects. Each one represents a unique challenge and solution.

Featured

Automatic Number Plate Recognition [ANPR] System

Engineered and optimized number plate recognition system for Bangladesh Road Transport Authority (BRTA) with FastAPI and model deployment.

PythonFastAPIComputer Vision

Vehicle Make-Model-Color [MMC] Recognition

Developed a framework for recognizing vehicle make, model, and color with accurate annotations.

PythonMachine LearningComputer Vision

Vision Language Model

Built infrastructure for multi-modal tasks like visual Q&A using vision-language prompts.

PythonVision-LanguageMultimodalAI

Featured

Py-Vision Platform

Led development of a scalable platform for CV microservices like ANPR, MMC, and Motion Detection.

PythonMicroservicesRedisObject Tracking

Featured

iVip Surveillance System

Integrated real-time analytics by combining ANPR, MMC, and VLM with Vision Relay.

PythonSurveillanceReal-timeAnalytics

Featured

Face Anti-Spoofing

Deployed a biometric anti-spoofing system for Bangladesh Election Commission.

PythonFace DetectionSecurity

Work Experience

My professional journey in the software development industry

Software Engineer

TigerIT Bangladesh Limited

2023 - Present

Current

Block-K, House#21 Rd No.28, Dhaka 1213

Lead developer for the company's multiple CV and AI microservices, including ANPR for BRTA using FastAPI, MMC vehicle recognition, Vision Language Model infrastructure, and real-time video analytics via iVip platform. Directed biometric security solutions like Face Anti-spoofing for the Bangladesh Election Commission, fine-tuned iris recognition systems, and implemented motion tracking for office monitoring. Designed dataset annotation tools and image generation systems for augmenting training data.

PythonArtificial IntelligenceMachine LearningComputer VisionDeep LearningReactRedisFastAPIMicroservicesMinIO

Machine Learning Intern

TigerIT Bangladesh Limited

2019 - 2021

Block-K, House#21 Rd No.28, Dhaka 1213

Contributed to speech recognition and NLP projects, focusing on voice verification and numeric normalization in TTS systems. Developed face filters with React JS to enhance user experience in company applications.

PythonNLPSpeech RecognitionReactJavaScript