Senior AI/ML Engineer, Data Scientist & Technical Lead with 4.5+ years of leadership experience, specializing in intelligent automation and enterprise AI deployment across international markets.

PROFESSIONAL_SUMMARY.log

Senior AI/ML Engineer and Data Scientist with 4.5+ years of leadership experience across international and local markets including Germany, and Bangladesh. Proven track record of scaling AI teams from 2 to 9 engineers while managing $800K+ R&D budgets and generating $2M+ business value through intelligent automation and enterprise AI deployment with extensive experience in on-site team management and stakeholder collaboration across enterprise environments.

Key Achievements:

  • Led cross-functional ML teams with 95% project success rate.
  • Deployed 50+ enterprise AI solutions (90%+ accuracy) in banking, logistics, surveillance, e-commerce.
  • Built scalable systems processing 82 live streams across 18 locations.
  • Implemented tobacco leaf SOP checker across 60+ locations with 13 ML models.
  • Published peer-reviewed papers and presented at international conferences.

Industry Impact:

Banking, FMCG, logistics, telecommunications, retail across Bangladesh, UAE, US, Germany, France, China, Australia with deep understanding of local market requirements.

EDUCATION_RECORDS.log

Master of Science in Computer Science and Engineering

BRAC University, School of Data and Sciences | Dhaka, Bangladesh

Thesis Ongoing | Percentage: 85-90% (Top 5%, High Distinction) - Merit-Based Scholarship

Relevant Coursework:

Symbolic Machine Learning A+ Advanced Syntactic Pattern Recognition A Neural Networks & Fuzzy Systems A- Mathematical Programming A- Data Science B+

Specialization:

Computer Vision, Pattern Recognition, Natural Language Processing

Supervisor: Dr. Md. Golam Rabiul Alam, PhD (Kyung Hee University Alumni)

Bachelor of Science in Computer Science

BRAC University, School of Data and Sciences | Dhaka, Bangladesh

Graduated 2019 | Percentage: 70-75%

Relevant Coursework:

Project & Thesis A Discrete Mathematics A- Data Structures A- Tensor Calculus A- Mathematical Physics A- General Theory of Relativity A- Machine Learning B+

Thesis: Machine Learning Approach to Credit Default Prediction achieving 89% accuracy

Supervisor: Dr. Mahbubul Alam Majumdar, PhD (Stanford, Cambridge, MIT, Imperial College Alumni)

PROFESSIONAL_EXPERIENCE.dat

Total Experience: 4.5+ years

Jul 2021 – Jun 2025 (2.5 years)

Bondstein Technologies Ltd. (Runner Group) | Dhaka, Bangladesh

Jul 2023 – Jun 2025 (2 years)

Team Lead, ML Research Engineer II

  • Led cross-functional AI/ML team of 9 engineers with 95% project success rate, managing sprint planning and strategic road maps.
  • Managed R&D budgets for AI infrastructure, processing units, and edge devices across multiple enterprise projects.
  • Involved in client meetings to analyze feasibility of the required features of their demanded products and designed long-term planning for product maintenance after deployment.
  • Monitored each team member’s task progress by maintaining sprints, mentoring one-to-one sessions, reviewed their codes before approving pull requests, provided resources and guidelines to my team for the new and old projects of this organization.
  • Developed and deployed scalable, real-time forecasting models for user behavior and sales on Azure Container Apps, achieving benchmark-low prediction errors.
  • Fine-tuned and integrated large vision-language models on Azure NC24A100 VMs to build advanced reasoning systems with computer vision, speech, and NLP capabilities.
  • Designed, optimized, and deployed custom AI monitoring system processing 82 live camera streams across 18 locations with 94.5% accuracy for theft detection.
  • Built comprehensive surveillance system for leading private bank featuring audio detection and violence recognition.
  • Implemented tobacco leaf SOP checker across 60+ locations using 13 ML custom models trained on 10K+ annotated images.
  • Developed hybrid NLP engine achieving 98% accuracy for data quality classification with automated retraining.
  • Created container tracking solution with Multimodal OCR approach with YOLO and PaddleOCR deployed on NVIDIA Jetson Nano & Nano Super using model optimization.
  • Architected multi-modal AI systems integrating LLAVA, LLAMA, and YOLO models achieving 92-97% accuracy for Azure edge deployment.
  • Delivered custom computer vision solutions for drone deployment with edge computing optimization.

Jan 2023 – Jun 2023 (6 months)

ML Research Engineer I

  • Built intelligent parking system achieving 95.47% number plate recognition and 97.12% occupancy detection accuracy.
  • Developed Driver Monitoring System with 94.61% DMS and 96.32% ADAS accuracy, presented at BASIS SoftExpo 2023.
  • Designed TV habit monitoring solution with 98.23% accuracy for consumer behavior analysis, deployed on edge device with model optimization and delivered as consumer-ready product.

Jul 2021 – Dec 2022 (1.5 years)

eaaasy (lookaround GmbH) | Munich, Germany (Remote)

Feb 2022 – Dec 2022 (11 months)

Lead AI Engineer

  • Led AI team of 5 engineers for MVP development, participating in strategic planning and technical recruitment.
  • Developed real-time brand logo detection using YOLOv5 achieving 96.71% accuracy for e-commerce applications.
  • Implemented acoustic fingerprinting algorithms for advertisement detection with custom audio processing.
  • Deployed ML models on AWS and GCP using Docker containerization and CI/CD pipelines.

Jul 2021 – Jan 2022 (7 months)

AI Engineer

  • Created four end-to-end recommendation systems using content-based and collaborative filtering approaches.
  • Deployed systems using Flask, Google Cloud Run, and Docker with comprehensive documentation.

Jan 2021 – Jun 2021 (6 months)

Ecomclips | Dhaka, Bangladesh

IT Project Manager & Business Intelligence Analyst

  • Managed multi-platform e-commerce projects across Shopify, WooCommerce, coordinating cross-functional teams.
  • Implemented analytics solutions using Tableau and Power BI for performance monitoring and sales forecasting.

TECHNICAL_SKILLS.exe

Programming:

Python Java R MATLAB SQL Bash

AI/ML Frameworks:

PyTorch TensorFlow Keras Scikit-learn Hugging Face YOLO LLAMA

Web Frameworks:

Django Flask FastAPI Streamlit

Computer Vision:

CNNs Transformers Paddle OCR LLAVA OpenCV Object Detection Image Classification Transfer Learning Knowledge Distillation

NLP & LLMs:

RNNs BERT GPT Claude Gemini spaCy NLTK Sentiment Analysis Recommendation Systems

Cloud & MLOps:

AWS GCP Azure Docker Kubernetes CI/CD mlflow Grafana Prometheus

Data & Analytics:

Pandas NumPy Matplotlib Seaborn Tableau Power BI

Time Series:

ARIMA ARIMAX SARIMA Prophet LSTM Random Forest Exponential Smoothing

Databases:

PostgreSQL MySQL MongoDB Redis nginx

Edge Computing:

ONNX Model Quantization Model Pruning NVIDIA Jetson Raspberry Pi On-premise Deployment

Operating Systems:

Ubuntu Windows Android Raspberry Pi OS Jetson Linux

MOST_INTERESTING_PROJECTS.log

Container Tracking Solution

// PROBLEM_STATEMENT

Track container IDs in real-time as they are moved by stackers and trucks in a private depot, identify their floor locations for pickup and drop-off, and update a live location on a customized depot map.

// SOLVING_STRATEGY

Set up cameras at gates and on stackers to record streams. Extracted frames to develop a custom YOLO model for real-time OCR of container IDs and movement detection. Deployed on edge devices connected via LAN, using a private network for data communication through custom APIs.

// ISSUES_FACED & RESOLUTION

Challenges included hectic on-premise deployments in a remote area, procurement of expensive edge devices, and model accuracy drops on smaller hardware. Solved by retraining models with more data, using model pruning, and arranging local accommodation for resources.

Portable Voice and Vision Assistant

// PROBLEM_STATEMENT

Create a device to explain images through its camera and provide prompt answers by listening to voice commands. It needed functionalities like map directions, weather updates, document scanning, and book reading, primarily for visually impaired users.

// SOLVING_STRATEGY

Utilized pretrained models from Hugging Face for vision and voice processing, deployed on an edge device. A custom API sends images to a VM for processing, and responses are sent back. Used multiple Docker containers for different APIs and a central VM for the database. Extended functionality to drones.

// ISSUES_FACED & RESOLUTION

The primary issue was internet dependency, causing slow responses. This was resolved by integrating 5G GSM support into the edge board for a faster, more reliable connection.

Deepfake for Wholesaler Campaign

// PROBLEM_STATEMENT

A celebrity needed to record a personalized 30-minute script for over 6,000 wholesalers on a tight 4-day schedule for a nationwide campaign, an impossible task with manual recording and editing.

// SOLVING_STRATEGY

Used GCP for Text-to-Speech generation of names, a voice cloning API to match the celebrity's specific Bangla dialect, and a lip-sync API for video generation. Scripted the process to crop and merge the generated clips into the original studio recording, automating the creation of 6,000+ personalized videos.

// ISSUES_FACED & RESOLUTION

Difficult name pronunciations and precise lip-sync timing were major hurdles. Overcame these through continuous iteration, quality monitoring by the team, and refining the video processing scripts to ensure seamless transitions.

DETAILS_OF_INDUSTRIAL_PROJECTS.dat

AI Surveillance for Financial Institution

Developed a comprehensive surveillance system featuring Bangla aggressive audio detection, custom object/face recognition, and violence detection integrated with real-time notifications for leading private bank.

Automated Lip Sync & Voice Cloning Campaign

Designed and deployed a complete pipeline leveraging multiple APIs and cloud resources to deliver over 6,600 personalized, lip-synced videos for a large-scale campaign within 4 days.

Tobacco Leaf SOP Checker

Architected a system with 13 custom classification/detection models trained on 10K+ annotated images, deployed across 60+ sites to ensure compliance with standard operating procedures.

Real-Time Face Recognition Monitoring

Developed a Siamese network with YOLOv8 for real-time face detection and recognition, optimized for extended operation in high-traffic environments with 94.5% accuracy.

Predictive Analytics for Production & Network

Implemented regression and ML models for production forecasting and network data analysis using time series techniques, delivering detailed performance reports with benchmark-low prediction errors.

Human Crowd Monitoring & Smart Parking

Built fusion models for crowd tracking and intelligent parking systems, achieving high detection accuracies of 97.12% for occupancy detection and 95.47% for number plate recognition.

Driver Monitoring System (DMS)

Developed advanced DMS with 94.61% accuracy and ADAS with 96.32% accuracy, optimized for edge deployment and presented at BASIS SoftExpo 2023.

TV Habit Monitoring Solution

Created consumer behavior analysis system with 98.23% accuracy for TV viewing patterns, deployed as edge-optimized consumer-ready product for market research applications.

NOTABLE_PERSONAL_PROJECTS.bin

Enterprise CCTV Face Extraction

Utilized Multi-Task Cascaded Convolutional Neural Networks (MTCNN) and OpenCV for high-accuracy face extraction from surveillance footage with optimized performance for enterprise environments.

ACCESS_SOURCE >

Healthcare EHR Classification Pipeline

Developed a robust pipeline for classifying patient severity states from Electronic Health Record (EHR) data using multiple ML and Deep Learning approaches with comprehensive evaluation metrics.

ACCESS_SOURCE >

Generative AI Image Systems

Built an Img2Img generator using Stable Diffusion with Diffusers from Hugging Face, exploring advanced generative capabilities and custom fine-tuning techniques.

ACCESS_SOURCE >

PDF Resume Data Extraction

Created an intelligent system for automatically parsing and extracting structured information from resumes in PDF format using NLP and document processing techniques.

ACCESS_SOURCE >

PUBLICATIONS_ARCHIVE.dat

Comparison of Multiple Machine Learning Approaches in Spam Detection

First Author - Springer Nature Book Chapter (ICACDS 2022)

Comprehensive comparison of ML algorithms for email spam detection with performance analysis and optimization techniques.

Implementation of Explainable AI in Mental Health Informatics

Co-Author - IEEE ICECE 2022

Research on implementing explainable AI techniques for mental health diagnosis and treatment recommendations.

Patients' Severity States Classification using ML & DL Approaches

First Author - arXiv:2209.14907 & GitHub

Novel approach to classify patient severity using multiple machine learning and deep learning techniques with EHR data.

A Machine Learning Approach to Credit Default Prediction

Undergraduate Thesis - BRAC University Digital Repository

Comprehensive study achieving 89% accuracy in credit default prediction using advanced ML algorithms and feature engineering.

A Survey: Implementations of NFT System in Different Fields

First Author - arXiv:2209.15288

Comprehensive survey of NFT implementations across various industries with technical analysis and future prospects.

AWARDS_AND_RECOGNITION.dat

Professional Recognition

2024

Best Team of the Quarter - Bondstein Technologies, R&D Department

2023

Best Employee of the Quarter - Bondstein Technologies, Technical Excellence

Academic Excellence

2022

Merit-Based Scholarship - BRAC University (Top 5% Academic Performance)

CERTIFICATIONS_AND_TRAINING.dat

AI & Business Strategy

Generative AI for Business Leaders AI & Business Strategy

Machine Learning & Data Science

Supervised Learning with Scikit-learn Time Series Analysis in Python Data Visualization with Seaborn