Hello there, I'm Avijit, Welcome to my personal website! I am a PhD candidate at the Manning College of Information And Computer Science, UMass Amherst. I am advised by Professor Hong Yu in the UMass BIONLP Lab. I completed both my BSc and MSc in Electrical and Electronic Engineering at Bangladesh University of Engineering and Technology (BUET).
With a strong foundation in both engineering and computer science, I've had the privilege of interning at Amazon Alexa AI for three consecutive summers (2021, 2022, 2023). My research now revolves around the development and practical implementation of Natural Language Processing (NLP) systems in the healthcare domain, with a specific focus on harnessing clinical data, including Electronic Health Record (EHR) notes.
Before embarking on my PhD journey, I contributed as a Machine Learning Researcher at Semion, where I dedicated my efforts to crafting and deploying AI-driven solutions in the realm of healthcare. My passion lies at the intersection of cutting-edge technology and its transformative impact on healthcare, and I am committed to advancing the field.
BSc in Electrical and Electronic Engineering, 2017
Bangladesh University of Engineering and Technology
MSc in Electrical and Electronic Engineering, 2021
Bangladesh University of Engineering and Technology
PhD in Computer Science, (Ongoing)
University of Massachusetts Amherst
A complete teleradiology solution that incorporates deep learning models to detect chest abnormalities. The framework is JavaFX and SQL is at backend. This was a professional project.
This is a kaggle competition where the objective was to segment salt image pixels from seismic images. With just two weeks of work, my team ended up being in the top 11% (355th out of 3234 teams). We used a modified U-net incorporating residual block that was trained at multiple stages using lovasz and focal loss functions.
This is an extension of my ongoing research work for deep learning based Sepsis screening system from EHR data. It's built using pyQt4. Here is a demo. This was a professional project.
This is the first kaggle competition to hold a large scale digit recognition challenge for bengali handwritten digits. My team became 12th out of 57 teams (top 23%). We used several image processing techniques and finetuned a pretrained densenet model.
This was performed on DRIVE dataset, using Unet based architecture with residual blocks. Fully connected neural networks and SVM were also used for comparative analysis.
To detect faulty semiconductor wafer from Japanese machine logs, the logs were first translated to English, went through multiple preprocessing steps and then a TextCNN based model was used. This was a professional project.
This is a hangman game but with a twist. This uses GRE words from Barron and Magoosh. Player can select difficulty level, word source, whether to show the meaning of a word in the end etc. I made this for fun.
Taking DWT features of an ECG signal, ANN, SVM and RF models were used to detect 4 categories of arrhythmia and a comparative study was performed. This was an academic project. The project report is available here.
This app was designed to help children stimulate their cognitive development. It uses an inception v3 model as it's backbone, pretrained on ImageNet dataset and fine-tuned on additional data. It was part of a research work which later got accepted in 2017 IEEE Region 10 Humanitarian Technology Conference (R10-HTC). Details are in the Publications section.
This Alexa skill can speak out all differential diagnoses for user fed symptoms. It has a comprehensive list of 1700 differential diagnoses and their corresponding mapping to symptoms.
Using bidirectional LSTM and CRF based entity recognition model, risk factors were detected for heart diseases in diabetic patients. The dataset was from i2b2. This was a professional project.
This is an android app that can detect diseases based on input symtoms. This also has a disease dictionary and an embedded google search link for each disease. The app is available in Google Play.
This was the final project for VLSI II Laboratory. This was an academic project.
This was the final project for Microprocessor and Interfacing Laboratory. This was an academic project.
A nokia 5510 display was interfaced with arduino, potentiometer and sonar sensors to build the whole system. Players had option to either use onboard switch for control or go free by giving hand gestures to move the bar. A demo is available on Youtube.This was an academic project.
Arduino along with GSM module, LCD panel, proximity and sonar sensors were used for this project. This was an academic project.
This was the final project for Digital Electronics Laboratory. This was an academic project.
This was a group project. We built a home automation system that can be controlled and monitored to switch and regulate electric loads using an android app we designed. This prototype used bluetooth to connect.