Summary
Machine Learning Engineer and applied researcher with 10 papers, 8 as first author (6 accepted or published, 4 under
review) in speech, computer vision and efficient ML. Takes models from research to production: runs self-hosted LLM
serving on in-house GPUs, builds speech systems (TTS, ASR, end-of-turn detection) for low-resource languages, and
designs lightweight models for real-time and resource-constrained deployment.
Education
B.Sc in Software Engineering, Department of CSE
June 2021 to September 2025 at Islamic University of Technology - CGPA: 3.80/4.00
- Thesis: Squeeze-Aware-YOLO: Early Channel Recalibration and Multi-Scale Fusion for Real-Time Traffic Sign Detection (accepted at DICTA 2026)
- English Proficiency: IELTS Academic, Overall Band 8.0
Work Experience
Machine Learning Engineer
October 2025 to Present at Vivasoft Limited, Banani, Dhaka
- Set up and maintain production-grade serving of self-hosted Gemma 4 models on in-house GPU servers using vLLM and SGLang, replacing external LLM API calls with local LLMs.
- Built a reverse-proxy gateway in front of the self-hosted LLMs that inspects request content, validates API keys, enforces rate limits, rotates TLS certificates automatically, and routes each request to the right model; monitored it with Prometheus and Grafana.
- Applied graph community detection to select community-aware core-sets for data-efficient TTS training (Structure Before Sampling, under review at ALTA 2026).
- Built and trained a custom Bangla TTS model on an in-house curated dataset (BanglaKontho, O-COCOSDA 2026).
- Developed a Bangla end-of-turn detection model for a real-time voice chatbot (BanglaTurn, O-COCOSDA 2026).
- Trained multiple Automatic Speech Recognition (ASR) models, including Whisper and NVIDIA NeMo, for the Nepali language.
- Built domain-specific Small Language Models (SLMs) optimized for fast CPU inference.
- Designed and implemented an efficient end-to-end Speech-to-Speech system for low-latency voice interactions.
- Designed and implemented an end-to-end ML data pipeline on GCP for data ingestion, preprocessing, and training ML models.
Associate Software Engineer
June 2024 to September 2024 at Vivasoft Limited, Banani, Dhaka
- Replaced a GPT-4o-based AI image search module with a self-hosted Vision Transformer model for an Indonesian e-commerce platform.
- Built a retrieval-augmented Shopify chatbot for product recommendation using GPT-3.5, LangChain, FastAPI and Pinecone Vector DB.
Software Developer Intern
April 2024 to May 2024 at Itransition Group, Remote
- Developed and deployed five Ruby on Rails applications with secure authentication and responsive Bootstrap interfaces.
Software Developer Intern
June 2023 to August 2023 at OpenGenus Foundation, Remote
Research Experience
Research Intern
March 2025 to Present at Advanced Machine Intelligence Research Lab, Remote
- Implemented federated learning and applied model pruning and quantization to improve efficiency and privacy in distributed training (PruneFedAD, DASA'25).
- Developed a lightweight classification model using depthwise convolutional networks, maintaining high accuracy for complex classification tasks.
- Enhanced skin cancer classification performance using ensemble techniques.
Publications
Accepted & Published
- Squeeze-Aware-YOLO: Early Channel Recalibration and Multi-Scale Fusion for Real-Time Traffic Sign Detection
28th International Conference on Digital Image Computing: Techniques and Applications (DICTA 2026). – First author
arXiv - BanglaKontho: Closing the Long-Form Gap in Bangla Text-to-Speech
29th International Conference of Oriental COCOSDA (O-COCOSDA 2026). – First author
arXiv - BanglaTurn: A Benchmark and Whisper-Based Model for End-of-Turn Detection in Bangla Speech
29th International Conference of Oriental COCOSDA (O-COCOSDA 2026). – First author
arXiv - Maximizing ChatGPT for Education: Strategies for Enhanced Learning Experiences
2026 2nd International Conference on Human-AI Interaction and Experience. – Second author
Google Scholar - PruneFedAD: Federated Learning with Structured Pruning for Low-Resource Alzheimer's Disease Detection
2025 International Conference on Decision Aid Sciences and Applications (DASA'25). – First author
Google Scholar - A Survey to Understand the experience of ChatGPT Usage among Engineering University Students in Bangladesh
Technium: Romanian Journal of Applied Sciences and Technology (ISSN: 2668-778X), 2023. – Second author
DOI
Under Review
- “It Looked Right to Me”: How the Artifact, Not the Practitioner, Sets the Limits of Verifying AI-Generated Work
ACM CHI Conference on Human Factors in Computing Systems (CHI 2027). – First author - Not Hearing or Not Negating? Disentangling Negation Failures of Audio-Language Models in English and Bangla
ACL Rolling Review (ARR), October 2026 cycle; to be committed to NAACL 2027. – First author - Structure Before Sampling: Community-Aware Core-Set Selection for Data-Efficient Text-to-Speech
arXiv preprint, 2026. Australasian Language Technology Association Workshop (ALTA 2026). – First author
arXiv - LeafLiteNet: A Computationally Frugal Interpretable Neural Architecture Under Controlled and Uncontrolled Environmental Conditions
Neural Computing and Applications (Springer), 2025. – First author
Achievements
- 2nd Runner Up: Streamstech presents IUT ICT Fest 2024 Code Catalyst: Refactoring Contest
- Finalist at AI and API Hackathon, United Group presents BUET CSE Fest 2024.
- Awarded AWS AI & ML Scholarship out of thousands of applicants across the world.
- Winner at ML Olympiad: Tensorflow User Group Bangladesh by Google.
- Placed one of top 5 teams in CodeRush 1.0 Hackathon.
- Secured a place in top 50 of the Huawei Seeds For The Future 2024, Bangladesh.
Extracurricular Activities
Microsoft Learn Student Ambassadors
- Hosted and organized educational workshops focused on web development and delivered engaging presentations particularly on HTML, CSS, JavaScript, Git, GitHub, Django, Azure, and Python.
Talks and sessions
AI Adda at Vivasoft Limited
- Conducted a knowledge-sharing session for software engineers on integrating AI into software applications, covering key concepts such as Transformer architectures, LSTMs, and Reinforcement Learning with Human Feedback (RLHF).
Get started with web development using Visual Studio Code and GitHub Pages
- Delivered an engaging session on web development using Visual Studio Code and GitHub Pages, guiding participants through the full-stack development workflow, including IDE usage, version control, and deployment practices.
Technical Skills
Languages: Python, C++, Go, Java, JavaScript
ML & Deep Learning: PyTorch, TensorFlow, NumPy, Pandas, Matplotlib, Seaborn
LLMs & Speech: vLLM, SGLang, LangChain, Whisper, NVIDIA NeMo
Cloud & Deployment: GCP, AWS (SageMaker, Lambda, Step Functions), Docker, Prometheus, Grafana
Databases: PostgreSQL, MySQL, Oracle DB, MongoDB, Redis, Pinecone
Web Development: FastAPI, Django, Node.js, Express.js, Ruby on Rails, React, HTML, CSS, Bootstrap
Developer Tools: Git and GitHub, VS Code, Postman, Selenium, JMeter