I completed my B.Tech. (Bachelors) in Computer Science from Indian Institute of Technology (IIT) Patna
where I spent some excellent four years of my life. My Bachelor Thesis supervised by Prof. Jimson Mathew received the Institute Proficiency Prize for Best Bachelor Thesis among students graduating in 2022.
During my bachelors, I worked as a Research Affiliate at MIT Media Lab
to create Brain-Computer Interfaces;
as a Software Engineer Intern at Sybill.ai;
collaborated with peers to work on automated early diagnosis of Alzheimer’s disease, and
have been a successful Google Summer of Code (GSoC) 2019 student contributing backend code to the Open Source organization Rocket.Chat.
Please feel free to check out my resume
and drop me an email
if you want to chat with me!
Indian Institute of Technology Patna Bachelor of Technology in Computer Science Engineering
July '18 - May '22
Awards: Institute Proficiency Prize for Best Bachelor Thesis
General Secretary (UGR) | Academic and Career Council, Students' Gymkhana
Coordinator | NJACK - Computer Science society
Member | Anwesha - Annual cultural festival
Research Fellow | Microsoft Research Lab India
Jul '22 - Present
Advisor: Dr Venkat Padmanabhan and Dr Akshay Nambi
My work spans across Systems and Networks, Machine Learning, and Optimization.
Research Affiliate | MIT Media Lab
Jul '21 - Nov '21'
Fluid Interfaces Group. Building the unsupervised machine learning and software engineering components to create
Brain-Computer Interfaces (BCI) for realtime-feedback in assistive devices.
Software Engineer Intern | Sybill.ai
Jun '21 - Aug '21
Working on the core infrastructure of a venture-backed, early-stage SaaS startup building an AI-powered video call
partner which provides contextualized insights on participants’ emotions. Currently building the user authorization
and asynchronous dataflow pipeline to communicate with Microsoft Graph API.
Google Summer of Code (GSoC) | Rocket.Chat
May '19 - Sep '19
Designed and developed "Newsfeed" (a social networking feature) for the Open Source application Rocket.Chat. Created a system to
retrieve messages from other channels and summarize them to prevent spam in real-time.
Worked on scalability and server performance using an adaptive follower/following model.
We present HyWay, short for "Hybrid Hallway", to enable mingling and informal
interactions among physical and virtual users, in casual spaces and settings,
such as office water cooler areas, conference hallways, trade show floors, and more.
Key to the design of HyWay is bridging the awareness gap between physical and virtual
users, and providing the virtual users the same agency as physical users.
*Author order reverse alphabetical
ScienceQA: A Novel Benchmark Resource for Question Answering on Scholarly Articles
Tanik Saikh, Tirthankar Ghosal, Amish Mittal, Asif Ekbal, Pushpak Bhattacharyya
International Journal of Digital Libraries (IJDL), 2022
We introduce a semi-automated dataset having more than 100k
human-annotated context-question-answer triplets to facilitate question answering
(QA) on scientific articles. Secondly, we implement
QA models based on BERT, SciBERT and a combination of
SciBERT and BiDAF and evaluate the results on our dataset. The
best model obtains an F1 score of 75.46%.
Multi-Modal Detection of Alzheimer's Disease from Speech and Text
Amish Mittal*, Sourav Sahoo*, Arnhav Datar*, Juned Kadiwala*, Hrithwik Shalu, Jimson Mathew
20th International Workshop on Data Mining in Bioinformatics (BIOKDD) with SIGKDD 2021
Reliable detection of the prodromal stages of Alzheimer's disease (AD) remains difficult even today because there is no definitive diagnosis of AD in vivo.
We propose a multimodal deep learning method that utilizes speech and the corresponding transcript simultaneously to detect AD.
We also perform experiments to analyze the model performance when ASR system generated transcripts are used and further perform an essential study of age and gender bias of our model. The proposed method achieves 85.3% 10-fold cross-validation accuracy on the Dementiabank Pitt corpus.
*Authors contributed equally
In collaboration with JCBC, University of Cambridge, UK
Making Gradient Descent non-monotonic over gradient (Bachelor Thesis)
Advisor: Prof Jimson Mathew
A novel non-monotonic gradient descent optimizer which aims to reduce the number of divergences while using gradient descent, along with its theoretical and empirical validation.
Decoding quantum states through nuclear magnetic resonance
Machine Learning for Physics. Built a model to predict the coupling parameters associated with nuclei and electrons
given their time-dependent magnetization curve from a spin-echo NMR experiment. Created a weighted ensemble model comprising of 1. Random Forest over statstical time-series features, 2.
InceptionTime, and 3. a custom CNN to achieve an R2 value of 0.992 and 0.997 for predictions on coupling strength and dissipation parameters respectively.
Developed an assembler and an emulator for a custom architecture machine
consisting of 2 registers, 1 program and stack counter, and some select mnemonics using C++.
The assembler can handle comments and data in different numerical bases and can detect multiple errors and warnings in the assembly code as expected.
The emulator loads the object file created from the assembler and can trace machine code and produce the memory dump along with identifying runtime errors.
Created the prototype of a functional multiplayer parkour video game using Unreal Engine 4; along with tutorial videos for it. Used State Machines to control behaviour and animations. It was my first attempt to learn about behaviour states and OOPs. Programming Language used - C++.
Last updated: Sep 1, 2023
This template is a modification to Jon Barron's website. It has further been modified by Rishab Khincha. Find the source code to my version here. Feel free to clone it for your own use while attributing the original author Jon Barron.