About
PhD researcher in Computer Engineering at UAH, working at the intersection of reinforcement learning, NLP, and intelligent transportation systems. I combine academic research and 4+ years of industry experience to build AI solutions for mobility, language, and real-world data.
Latest News
Stay updated with my latest activities, research, or publications.
Paper Accepted at IEE ITSC 2025
Published my paper on the data-driven modeling of the car-following behavior for Electric Vehicle at IEEE ITSC 2025, Gold coast, Australia.
Presented Presented at IEEE UEMCON 2024
Published my paper on "Bangla SBERT" at IEEE UEMCON 2024 held in New York.
Submitted paper at IEEE MLCAD 2025 Symposium
Worked on CPU Burst Forecasting and improve the latency using ML
Education
PhD in Computer Engineering
2025 - Present
University of Alabama in Huntsville (UAH), USA
Research Focus: Reinforcement Learning, NLP, and Intelligent Transportation Systems
Master Of Science in Information and Communication Technology
2021 - 2022
Comilla University, Comilla
CGPA : 3.54/4.00
Bachelor Of Science in Information and Communication Technology
2016 - 2020
Comilla University, Comilla
CGPA : 3.70/4.00
Publications
1. Modeling Electric Vehicle Car-Following Behavior: Classical vs Machine Learning Approach , accepted at IEEE ITSC 2025
Shihab Uddin, Md. Nazmus Shakib, Rahul Bhadani.
DOI (pending or not yet public)
This comparative study evaluates classical and machine learning approaches for modeling electric vehicle car-following behavior, providing insights into the strengths and limitations of each methodology in capturing EV-specific driving dynamics.
2. Bangla SBERT - Sentence Embedding Using Multilingual Knowledge Distillation
M. S. Uddin, M. A. Haque, R. H. Rifat, M. Kamal, K. D. Gupta and R. George.
DOI: 10.1109/UEMCON62879.2024.10754765
This paper presents Bangla SBERT, a distilled sentence embedding model optimized for the Bangla language. We apply multilingual knowledge distillation to enhance semantic understanding for low-resource NLP tasks.
3. Semantic Topic Extraction from Bangla News Corpus Using LDA and BERT-LDA
P. C. Paul, Md. Shihab Uddin, M. T. Ahmed, M. Moshiul Hoque and M. Rahman.
DOI: 10.1109/ICCIT57492.2022.10055173
This study explores topic modeling in Bangla using LDA and BERT-LDA, demonstrating the improved coherence and semantic accuracy of transformer-enhanced topic extraction from large-scale news corpora.
4. A Cross-Domain Exploration of Audio and Textual Data for Multi-Modal Emotion Detection
Ariful Haque, M., George, R., Hossain Rifat, R., Md. Shihab Uddin, M. Kamal, & Datta Gupta, K.
This work investigates multi-modal emotion detection using both audio and text inputs. The proposed models reveal that domain-specific training significantly improves emotion classification across modalities.
5. Efficient CPU Burst Forecasting with Low-Latency Machine Learning Models (under review, IEEE MLCAD 2025)
Md Nazmus Shakib, Md Ashraf Hossain Ifty, Md. Shihab Uddin, Rahul Bhadani.
This paper investigates the use of ML models (KNN, DT, RF, MLP) for predicting CPU burst times based solely on submission-time attributes. These models are benchmarked against the exponential averaging (EA) method using the GWA-T-4 AuverGrid dataset. The study evaluates not only prediction accuracy but also inference latency for real-time scheduling. While KNN achieves high accuracy, the DT model shows the best trade-off between accuracy (MAE 3514.09, CC 0.84) and latency (0.0033 ms/sample). Confidence intervals and robustness testing confirm the feasibility of using ML for real-world CPU scheduling.
Skills
Professional Experience
Senior Software Engineer
Silicon Orchard Ltd.July 2023 - January 2025
- Designed Proof of Concepts (POCs), and Software Workflows to address specific business challenges.
- Managed the project with cross-functional teams and clients to design, implement, and deploy features.
- Designed and developed property recommendation system,property savings prediction model.
- Developed APIs to serve ML models using frameworks like Flask, Fast API.
- Leveraged Large Language Models (LLMs) to tackle tax applications while reducing cost & response time.
Taxation System Optimization
- Perform thorough data preprocessing and augmentation for 3D medical image datasets.
- Construct and optimize 3D image classification models for medical data with a focus on interpretability
- Implement IBA and GradCam techniques to enhance the interpretability of the 3D medical image classification models
Explainable 3D Medical Image Classification
- Create novel datasets by merging diverse open-source datasets for both audio and text modalities.
- Build accurate audio and text classification models with a target accuracy of 84% and 73%, respectively, using advanced sequence learning techniques
- Construct an efficient inference engine using Streamlit for real-time sentiment detection in both audio and text modalities.
Double Modality Sentiment Detection
Software Engineer (NLP)
Technometrics LimitedJune 2022 - June 2023
- Built a text data pipeline which reduce the process time 45%
- Using a data-centric approach, i created a Bangla language modeland increased its accuracy by 82% over previous training.
- Reduced end-to-end ASR training time by 77% by utilizing bucketing technique.
- Reduced end-to-end ASR model Word Error Rate from 31 to 25 by utilizing self training / semi-supervised techniques for ASR.
- Built an automatic speech data collection pipeline and collected 20 thousand hours of speech data from the public internet.
- Created a 30 thousand hours dataset of Speech data using data augmentation of 15 thousand hours.
- Developed, and maintained an automatic cookies-based login system for an automation platform.
- Created APIs for different statistical features for the system
Bangla Language Model
Bangla ASR
Social Media Monitoring System
Junior Machine Learning
Ishraak Solutions LimitedDec 2020 - May 2022
- Developed a resume extractor which can extract user information from their resume using BERT.
- Developed an NLP β based job extractor that extracts information from unstructured job data.
- Developed an application to detect job posts or not with accuracy of 98% Worked on an application that verifies user profiles to see if the profile photo is appropriate or not.
- Worked on a hybrid Job Recommendation.
- β Worked on a job categorization application that classified 50 categories with 81% of accuracy.
- Worked on Multi Classification and Binary Classification problems based on NLP.
- Developed a Custom Named Entity Recognition Using Spacy
JobXprss
Extra Caricular Activites
President, Comilla University IT Soceity
2020 - 2021
Joint Secretary, ICT Association, Dept. Of ICT
2019 - 2020
Radio Jockey, RadioCou
2018 - 2020
Language Proficiency
English, Overall 6.5
- Speaking 7.0
- Reading 6.5
- Listening 6.5
- Writing 6.0
Bangla, Native
Testimonials
Contact
Location:
Huntsville, Alabama, USA
Email:
shihabuddin.ict@gmail.com