K. M. Asifur Rahman

AI/ML Engineer | Researcher

Engineering Intelligence at the Intersection of AI, Security, and Innovation.

Contact Research Projects CV

About Me

I’m a Machine Learning Engineer and Cybersecurity enthusiast who loves building intelligent systems that are not only powerful but also secure. My journey spans GenAI, Computer Vision, and NLP, where I’ve explored mostly from fine-tuning LLMs to detecting deepfakes and strengthening 5G security. Along the way, I’ve had the chance to publish a research on 5G vulnerabilities and deploy AI models in highly restricted, bank-grade environments—where every detail mattered for both performance and security. What excites me most is bridging the worlds of AI innovation and cybersecurity resilience, turning cutting-edge ideas into real-world solutions.

Experience

July 2024 - Present

Machine Learning Engineer

Spectrum Software and Consulting(Pvt) Limited, Karwan Bazar, Dhaka

Built an AI-powered eKYC platform for identity verification using liveness detection and face verification. The system validates user head movements against dynamic instructions, detects replay attacks from screens, and matches live facial images with National ID records. Optimized deployment on NVIDIA DGX Spark and NVIDIA A5000 GPU servers using stateless APIs and Gunicorn workers to improve memory efficiency, scalability, and inference latency.
Developed and deployed Doc-Ai systems that can extract data from both structured and unstruc tured documents and take necessary decisions or measures based on requirements. Deployed in an asynchronous scalable design for handling a large number of users.
Built a Knowledge Graph-RAG-based document agent, customized fully on local projects and documents. The agent can answer user queries from these artifacts. Reduced hallucinations by establishing relationships between documents and their context.
Researched on MCP, Large Language models, LLM Inference Engines, Lora, and Qlora finetunings. Applied fine-tuning and advanced memory management techniques in model customization and deployment.
Experienced in developing computer vision and voice-enabled smart systems that can recognize persons and summarize meetings.
Nov 2023 - Jan 2024

Intern

CIRT, Bangladesh Computer Council

Explored an open source SIEM (Wazuh) to deploy, customizing it for a few features from an AI/ML perspective. Tried to integrate some ML models for anomaly detection from network data and logs. Worked with cyber experts, had a great learning experience.
May 2023 - Jun 2023

Intern

Era InfoTech Ltd, Purana Paltan, Dhaka

Developed a transaction classifier using Natural Language Processing (NLP) tools, clustering algorithm and Python libraries. Elevated data analysis precision and improved model accuracy through effective collaboration with teams.
May 2023 - June 2024

Director of Logistics

BUET Cyber Security Club

Successfully organized and executed a Capture The Flag (CTF) competition, showcasing adept coordination, leadership, and problem-solving skills.

Research

5GPT: 5G Vulnerability Detection by Combining Zero-Shot Capabilities of GPT-4 with Domain Aware Strategies Through Prompt Engineering

Accepted in IEEE-TIFS

In this research I worked under the guidance of Prof. Md. Shohrab Hossain, alongside Prof. Ren-Hung Hwang and Prof. Ying-Dar Lin from NYCU, Taiwan, on exploring the vulnerabilities of 5G. Our goal was to develop a system capable of identifying security weaknesses in 5G specifications. This work led to our paper, ”5GPT: 5G Vulnerability Detection by Combining Zero-Shot Reasoning and Domain-Aware Prompting,” which has been accepted for publication in IEEE-TIFS. Our approach combined GPT-4’s zero-shot reasoning capabilities with a domain-aware prompting strategy, where we incorporated telecom-specific security properties, signaling rules, and hazard indicators. This allowed the model to generate more accurate and technically grounded vulnerability hypotheses compared to standard prompting.
My responsibility was to put our ideas to the test; I dug deep into the UERANSIM codebase. I tried to trace each potential vulnerability back to exact parts of the user equipment (UE) and core network implementations. Having that code-level insight helped me simulate tricky scenarios like malformed messages (sent altered hash in reply), rare protocol deviations (turned on/off ACK messages), and unusual signaling patterns (sent random messages while the registration process was going on) with great precision. With our collaborative effort, this combined approach of smart AI prompting and hands-on system analysis uncovered 47 potential vulnerabilities, including 27 that had not been reported before. We were able to confirm 9 of those through practical testing using Open5GS and UERANSIM.

Bypassing Conventional eKYC: How Far Can We Go using Deepfake?

Accepted in 12th International Conference on Next Generation Computing, Communication, Systems and Security. Won The Best Paper Award, NSysS-2025

This research explores both the creation and detection of deepfake content to better understand the dual use of generative AI in multimedia. On the generation side, it implements face-swapping pipelines using InsightFace, ONNX models, Variational Autoencoders (VAEs), and prototype GAN architectures to create realistic manipulated videos. On the detection side, it fine-tunes state-of-the-art transformer models (ViT, SigLIP) and integrates an LSTM classifier to capture temporal inconsistencies across video frames. The framework is optimized for GPU acceleration on Google Colab, supports automated dataset preparation, and offers reproducible pipelines for training, evaluation, and inference. By combining generation and detection in a single research environment, this work provides insights into both adversarial capabilities and defensive strategies against synthetic media threats.

5G RAG Based Conformance Testing

On Going

We proposed a fully automated end-to-end framework that utilizes a Retrieval-Augmented Generation (RAG) pipeline. Our approach grounds LLM outputs in verified, domain-specific data to minimize hallucinations, and overcomes cross-section dependency challenges by integrating a robust con- text retrieval mechanism. Using this approach, we have generated 800 conformance test-cases for essential 5G mobility management procedures in under 3 hours. Overall, our framework offers a scalable, reliable, and robust solution for automating compliance testing in complex, ever-changing domains like 5G and beyond.

Data Leakage Detection in Microservices

On Going

I am working on Data Leakage Detection in Microservices, under the supervision of Prof. Md. Shohrab Hossain from BUET and Prof. Suryadipta Majumdar from Concordia University, with the support from United International University (UIU). This funded research investigates data leakage threats in containerized cloud environments by analyzing malicious container images and runtime attack techniques in Docker and Kubernetes. We simulate adversarial scenarios including reverse-shell execution and covert secret exfiltration to evaluate the effectiveness of runtime monitoring, syscall tracing, and container security mechanisms. To strengthen container security, we evaluate existing vulnerability scanners and runtime detection approaches, highlighting their limitations in identifying malicious behavior that does not rely on known software vulnerabilities.

An AI-Driven Audio-to-Report Framework to Automate Damage and Loss Assessment for Disaster Management

International Conference on Applied Statistics and Data Science 2025

This study presents an AI-driven audio-to-report system that generates structured climate loss and damage reports from Bangla dialect speech, targeting disaster-affected, low-literacy communities. By integrating speech recognition, NLP, and automated reporting, the framework demonstrates strong performance across multiple Bangladeshi dialects, highlighting its potential as a scalable and efficient solution for rapid climate disaster reporting.

Assessing the Impact of Temperature Change on Discomfort Index: Temporal Trends and Seasonal Variations in Dhaka City

ICWFM 2025

This study analyzed the temporal trends and seasonal variations in maximum temperature, minimum temperature, average temperature, and discomfort index in Dhaka city from 1981 to 2020.The discomfort index calculated using Thom Discomfort Index equation and employs a combination of statistical techniques to assess the impact of temperature changes on the discomfort index.

Exploring Post-Mortem Neural Signal Processing: Uncovering Computational Potentials in Deceased Animal Brains

Student Poster Champion NSysS 2021

We investigate the potential of a deceased animal brain to process signals. Specifically, we examine the brain’s responses to external stimuli in the form of electrical signals and its ability to act as a memory unit. We also explore the transfer characteristics of the deceased goat brain and elucidate the corresponding function through representative circuits.

Grants & Funding

Improving Computer and Software Engineering Tertiary Education Project (ICSETEP)

Funding Agency: University Grants Commission (UGC), Bangladesh

Project Title: AI-Powered Contradiction and Conflict Detection in Evolving Documents
Role: Researcher

Duration: Dec 2025 - Dec 2027

Focus: RAG and knowledge graph based document understanding and contradiction detection over time.

Skills for Industry Competitiveness and Innovation Program (SICIP)

Funding Agency: Bangladesh Industry Research Development & Innovation (BIRDI)

Project Title: AI-Augmented Repository-Level Bug Localization and Automated Patch Generation System with Large-Scale Codebase Dataset
Role: Researcher

Duration: Under Review

Focus: Automated software debugging and improvement of code reliability using AI-driven techniques, produces a high-quality debugging dataset, strengthens AI-based software engineering capabilities.

Projects

MeshMITM – Simulating MITM in Kubernetes Service Mesh

This project simulates Man-in-the-Middle (MITM) attacks in a Kubernetes + Istio service mesh and builds an anomaly detection framework based on latency features. Started with simuating various MITM attacks (service impersonation, traffic interception, metadata exposure) in a microservice architecture, and mitigation using Istio AuthorizationPolicies and scoped service accounts. Then developed an anomaly detection framework using Prometheus metrics to identify MITM-induced anomalies. By analyzing p50/p90/p99 latencies, dispersion indices, burstiness, drift measures, the system captures subtle delays and artifacts caused by malicious proxies. Lightweight models such as One-Class SVM, Isolation Forest, and shallow autoencoders are applied with sliding-window post-processing for stable detection.

Tech Stack:

Kubernetes (Minikube) Istio Python Numpy Pandas Flask Docker mTLS Microservices
View Details

Climate Report generation from Bengla Audio

Finalist at Ai Meet Climate Actions, OXFAM

The Audio Reporting System is an AI-powered platform that transforms disaster-related phone calls into structured, actionable PDF reports for emergency response teams. Designed to work without smartphones, apps, or internet access, it captures voice reports, transcribes them into text, extracts key details, and generates professional reports enriched with contextual information. With built-in scalability, human validation options, and a fine-tuning pipeline for continuous improvement, the system ensures faster, more reliable crisis reporting while empowering responders to act quickly and effectively.

Tech Stack:

Facebook M4tv2 model Gemini 2.0 Pydantic Pandas ML models Flask Docker RabbitMQ MongoDB Ml Flow
View Details

Deepfakes

Best Paper Award, NSysS-2025

This project investigates generating and detecting deepfake content to examine the dual use of generative AI in multimedia. It uses face-swapping with InsightFace, VAEs, and GANs for realistic video edits, while fine-tuning transformer models (ViT, SigLIP) and an LSTM classifier for temporal detection. Optimized for GPU on Google Colab with automated dataset management, it offers reproducible pipelines for training, evaluation, and inference, providing insights into both deepfake creation and defense

Tech Stack:

PyTorch Torchvision Transformers InsightFace ONNXRuntime-GPU OpenCV Pillow Imbalanced-learn SigLIP-2 LSTM
View Details

Neer–Automated Quality Assurance and Monitoring of Urban Water Usage

Finalist at WICC, 2022

This project aims to design a smart water management model aligned with the under-development District Metering Area (DMA) System. The proposed system leverages IoT-enabled digital meters to automatically update water consumption and quality data to secure cloud storage at regular intervals. IoT sensors will continuously measure various water quality parameters, ensuring real-time monitoring and transparency. The stored data will be systematically fed into a machine learning model, enabling advanced comparison, analysis, and forecasting of water quality trends across different regions. By integrating IoT, cloud computing, and predictive analytics, this system will support proactive water management, early anomaly detection, and sustainable resource utilization.

Tech Stack:

Arduino IoT sensors Wi-Fi Cloud database ML models
View Details

Anomaly Detection in ECG

Academic Project (CSE-472)

Project uses transformer model (encoder) layer to separate healthy heart signals from unhealthy ones. We used MIT-BIH Arrhythmia dataset to train and test our model. The model achieved high accuracy in classifying ECG signals. The project also includes data visualization using libraries like Matplotlib and Plotly to analyze ECG patterns and anomalies.

Tech Stack:

NumPy Pandas Matplotlib Plotly PyTorch Transformer Python
View Details

Gender Inequality and Climate Change Analysis

Nasa Space Apps Challenge, 2024

This project explores the relationship between gender inequality and climate change. The analysis leverages various data visualization and data manipulation libraries in Python to provide insights into how these two critical issues intersect. This was crafted to participate in the Nasa Space Apps Challenge, 2024 competition.

Tech Stack:

NumPy Pandas Matplotlib Plotly PyTorch Transformer Python
View Details

Metal Trade Clustering Analysis

Collaboration Project

Collaborated with the Mechanical Engineering Department on a research project analyzing raw material import data (2017–2021). Applied K-Means and Gaussian Mixture Models to cluster metals based on import quantities and price variations

Tech Stack:

Python Pandas Numpy Scikit-learn Matplotlib K-means clustering GMM clustering
View Details

Dynamic Document Parser

Academic Project (CSE-408)

Software designed to automate document processing by leveraging OCR technology and dynamic document templates.

Tech Stack:

Git CI/CD Docker Microservices Tesseract MongoDB Prisma EJS JavaScript
View Details

Peek a Book

Academic Project (CSE-216)

PeekAbook is an online book buying site where one can find various bookshops and buy books from those shops.

Tech Stack:

OracleDB PL/SQL Handlebars REST API HTML CSS
View Details

3D Solar System

Java application utilizes JavaFX to create a visually engaging 3D model of the solar system.

Tech Stack:

Java JavaFX Threading 3D Animation
View Details

Technology

Network and Security Tools

Wireshark Open5gs UERANSIM NS2 Nmap Wazuh Blackeye MaskPhish SRSran

Machine Learning & AI

TensorFlow PyTorch Scikit-learn SciPy Hugging Face LangChain OpenAI Gemini KNN GMM DBSCAN Hierarchical Clustering

Computer Vision & NLP

OpenCV Tesseract EasyOCR PaddleOCR InsightFace DeepFace VGGFaceNet NLTK Transformers

Data Science & Visualization

Pandas NumPy Matplotlib Seaborn FAISS Pinecone Vector DBs neo4j

LLM Engines & MLOps

Ollama VLLM TGI Lorax MLflow Docker Apisix Firebase MinIO RabbitMQ

Programming & Databases

Python Java SQL JavaScript Bash C++ C MongoDB Atlas Oracle MySQL SQLite Cypher

Education

BSc in Computer Science and Engineering

Bangladesh University of Engineering and Technology (BUET)

CGPA: 3.55/4.00 (Senior year CGPA: 3.70/4.00)

College

Notre Dame College, Dhaka

GPA: 5.00/5.00

Online Courses

  • Agentic Knowledge Graph Construction with neo4j (Deeplearning.ai)
  • Neural networks and Deep learning, Improving Deep Neural networks: Hyper parameter tuning. (Coursera)
  • Intermediate Machine Learning (Kaggle)
  • Python for everybody (5 course specialization) (Coursera)
View Certificates

Awards & Achievements

Best Paper Award

Won The Best Paper Award at the 12th International Conference on Next Generation Computing, Communication, Systems and Security. Paper Title: Bypassing Conventional eKYC: How Far Can We Go using Deepfake?

2025

Poster Presentation Champion

Became Poster Presentation Champion at International Conference on Networking System and Security (NSysS)

2021

Finalist

AI Meets Climate Action by OXFAM

2025

Finalist

WICC (Water Innovation Challenge Competition)

2022

Scholarship

Government scholarship awarded by Dhaka Education Board for brilliant academic performance in Higher Secondary Certificate public examination.

2018

Scholarship

Government scholarship awarded by Rajshahi Education Board for brilliant academic performance in School Secondary Certificate public examination.

2016

বিভাগীয় পর্যায়ে সেরা মেধাবী (Divisional Champion, Rajshahi division)

সৃজনশীল মেধা অন্বেষণ প্রতিযোগিতা, বিজ্ঞান (Creative Talent Hunt, Science)

2013

Sports

Inter Faculty Cricket Tournament Runner-up

2023

Contact Me

Dhaka, Bangladesh

(+88) 01882637342

asifurbuet98@gmail.com

asifurndc8030@gmail.com