Research Interest

Natural Language Processing, Multimodal Learning, Large Language Models, Vision Language Models

Education

Bachelor of Science (B.Sc.) in Computer Science and Engineering
- University of Dhaka (January 2017 – August 2021)
- CGPA: 3.74 out of 4.00

Therap (BD) Ltd.
- Machine Learning Engineer (October 2022 – Present)
  - Developed a computer vision-based product for remote patient monitoring within medical care facilities.
  - Led the development of a video-redaction system, enabling precise redaction of target individuals using Person Detection, Instance Segmentation models, and Multi-object Trackers.
  - Designed stream-processing pipelines with optimized computer vision models using NVIDIA Deepstream SDK, enabling real-time video analytics on NVIDIA Jetson devices.
  - Contributed to a U.S. patent for a non-invasive, real-time identification and redaction system in a monitored environment.
- Associate Machine Learning Engineer (September 2021 – September 2022)
  - Explored deep learning-based Computer Vision tasks like Object Detection, Face Recognition, Pose Estimation, Segmentation, and Activity Recognition for different scenarios.
  - Experimented with cutting-edge Activity Recognition models and assessed their real-time performance on live camera feeds.
  - Worked with depth sensors to evaluate the effectiveness of depth-map-based Human Activity Recognition models.

Center for Computational & Data Sciences (CCDS), Dhaka, Bangladesh
- Research Assistant (Part-Time) (September 2023 – Present)
  - Evaluated the performance of Large Language Models (LLMs) on low-resource languages like Bangla using techniques like prompting and fine-tuning.
  - Pretrained a new Bangla Language Model, Bangla-Llama-2-7B, based on Meta’s Llama-2 model with a 12 GB Bangla corpus using LORA methodology. Code, HF Hub
  - Currently working on training instruction-following LLaMA-based models for Bangla, aiming to enhance natural language understanding and generation in a low-resource setting.
Artificial Intelligence and Cybernetics Lab (AGenCy Lab), Dhaka, Bangladesh
- Research Assistant (Part-Time) (February 2022 – October 2022)
  - Constructed a heterogeneous biological network with genes, diseases, drugs, and biological functions from eight public biological databases.
  - Developed a representation learning approach for predicting novel Disease-Drug and Disease-Gene associations.

Faria Sultana, Md Tahmid Hasan Fuad, Md Fahim, Rahat Rizvi Rahman, Meheraj Hossain, M Ashraful Amin, A K M Mahbubur Rahman, Amin Ahsan Ali, How Good are LM and LLMs in Bangla Newspaper Article Summarization, to appear in the Proceedings of the 27th International Conference on Pattern Recognition, ICPR 2024. Paper
Md Fahim, Meheraj Hossain, Sadman Rohan, Md Ashraful Amin, AKM Mahabubur Rahman, Amin Ahsan Ali, L-Context: Layer-wise Context Vectors for Better Text Classification Using Pre-trained Language Models, under review. Paper

David Lawrence Turock, Justin Mark Brockie, James Michael Kelly, Richard Allen Robbins, Meheraj Hossain, et al., Automated, Non-Invasive Artificial Intelligence Machine Learning Method and System for Identifying and Redacting Personally Identifiable Information in a Monitored Environment using Real-Time Sensor Data, US Patent Publication No. US 2024-0212804 A1, published June 27, 2024. Patent

Secured 5th Position in Apurba Presents Bhashabhrom: Bangla Grammatical Error Detection Challenge Datathon 2023 (Team: Team Aambella). Link
Selected as Finalist in Robi Datathon 2.0 (Team: The_Anomalies). Link
Awarded University Merit Scholarship by the Government of Bangladesh for outstanding academic performance.

Undergrad Thesis | Machine Learning, Feature Selection, Bioinformatics, Data Mining (April 2021)
Title: mMultiSURF – A relief-based feature selection method considering class overlapping areas among neighboring instances and prior information. Thesis Book, Code
- Enhanced the MultiSURF algorithm to improve robustness and accuracy in selecting relevant feature subsets within high-dimensional datasets.
- Incorporated an instance weighting method, reflecting the likelihood of non-overlapping regions in calculating feature importance.
Amar Health | HTML, CSS, Node.js, Express.js, MongoDB (January 2020)
Developed a web-based application for electronic health record management and patient monitoring at “Telepsychiatry Research and Innovation Network Ltd” in Dhaka, facilitating psychiatric care and research. Code
ML Algo Simulator | HTML, CSS, Python, Flask, Machine Learning (April 2019)
Developed a web application to simulate basic machine learning algorithms on sample datasets. Code
MedAdvisor | Java, Android, SQLite (December 2018)
Created an Android app that provides users with information on various diseases, assisting in diagnosis and medication recommendations. Incorporated a medication reminder calendar for tracking doses. Code

Programming Languages: Python, C, C++, Java, JavaScript
Libraries: PyTorch, PyTorch-Lightning, TensorFlow, Keras, OpenCV, Scikit-learn, Numpy, Pandas, Matplotlib, Seaborn
Frontend Development: HTML, CSS, Bootstrap, jQuery, Ajax
Backend Development: Node.js, Express.js
Database: MongoDB, SQL, SQLite
Hardware Tools: Nvidia Jetson Xavier NX, Jetson AGX Orin, Jetson Orin Nano, Raspberry Pi
Miscellaneous: Git, Docker, MATLAB, LaTeX, TensorRT

Competitive Programming
- Solved 1000+ problems on platforms like Codeforces (Max Rating: 1527), LightOJ, and UVA.
- Participated in several national and international programming contests during undergraduate studies.
Kaggle Competitions
- Participated in competitions such as Google Brain Ventilator Pressure Prediction (Time Series Analysis) and Global Wheat Detection (Computer Vision Challenge).