Akshat Sanghvi Research Experience

B.Tech CSE (Honours) at IIIT Hyderabad 7/23‐NOW Honours (CVIT) Center for Visual Information Technology
CGPA: 9.01 till 7th Sem
LinkedIn Github Personal Website • Personalized Lip‐Reading for Deaf Speakers: Customized pre‐
Email: [email protected] trained Visual Speech Recognition (VSR) models to improve lip‐
Contact: +91‐6352927215 reading performance for out‐of‐distribution Deaf (and accented)
English speakers. Curated a dataset featuring speakers with un‐
Synopsis clear or no speech to enhance lip‐reading accessibility for the Deaf
Final (4th) year B.Tech student in Com‐ community. (Submitted work currently under review.)
puter Science at the International In‐
• Visual Question Answering (VQA) Web App: Designed and built
stitute of Information Technology Hy‐
derabad (IIIT‐H), pursuing an honours
web applications to showcase the answering capabilities of differ‐
degree with a focus on research. Cur‐ ent VQA models in diverse domains, including Medical, Road, and
rently working in three research labs Document VQA.
here at IIIT‐H : CVIT (Computer Vision), Advised by Dr. Jawahar C.V and Dr. Vinay P. Namboodiri
RRC (Robotics) and MLL (3D vision). En‐
10/23‐NOW Independent Study (RRC) Robotics Research Center
joy playing chess a lot, and served as
a coordinator of my college chess club • Critical Object Estimation for Self‐Driving Cars : Designing and im‐
and have a FIDE rating of 1730. plementing algorithms for real‐time path planning in autonomous
vehicles, optimizing for computational efficiency by prioritizing es‐
Skills sential vehicle interactions, unlike traditional methods that are lim‐
ited to analyzing a fixed number of closest vehicles.
LANGUAGES
Python, C, C++, JavaScript, Advised by Dr. K. Madhava Krishna and Dr. Arun K. Singh
HTML/CSS, Bash, x86, cuda 7/24‐NOW Independent Study (MLL) Machine Learning Lab
ML FRAMEWORKS • Compact 3D Scene Representation Developing methods for 3D
PyTorch, TensorFlow, Numpy scene reconstruction and novel view synthesis using Gaussian
OTHERS Splatting, addressing challenge of large model sizes (up to a giga‐
Markdown, Git, Vim, React.js, byte) for extensive scenes. Leveraging local repetitions and sym‐
Flask, Node.js, MySQL, MongoDB metries to achieve significant storage compression without com‐
promising quality.
Achievements Advised by Dr. Avinash Sharma and Dr. Charu Sharma
2023 Merit List (Monsoon)
2023 Merit List (Spring) Publications
2021 Deans List 2 (Monsoon)
2021 KYPY Rank 186 SEPT. 2024 DeafVSR: Personalizing Lip Reading for Deaf Speakers
2020 RMO Qualified This work presents a personalized approach to automatic lip reading,
2019 NTSE Stage 2 Qualified considered to be one of the most important assistive technologies for
2021 JEE Mains Rank 156 the deaf community. Employed layer‐specific fine‐tuning to identify
2021 JEE Advanced Rank 2019 the most effective parameters in the pre‐trained model for speaker‐
specific learning. The work is submitted to the prestigious ICASSP
Coursework conference and is currently under review.
Computer Vision
Mobile Robotics Projects
Linear Algebra
Digital Image Processing 2023 Image‐Space Manipulation of Objects in Video CV, ML
Advanced NLP Created 2D simulations of object movement in response to virtual
Statistical Methods in AI forces, analyzing video of tiny motions to infer material properties
Data Structures and Algorithms via modal analysis and a spring‐based physics model, predicting pixel
Operating Systems and Networks reactions to user‐defined forces.
Information Security
Quantum Information and
2024 Exemplar Guided Paraphrase Generation NLP, ML
Computation Developed ML models for paraphrase generation that uses example
Data and Applications sentences (exemplars) to guide rephrasing while preserving the ori‐
ginal meaning, utilising the concept of contrastive loss on the style
feature and content features of the text
Education Projects

B.TECH. IN CSE (2021‐NOW) 2023 GMM Visualization Manim Web, Statistical Methods in AI
IIIT ‐ Hyderabad Created a comprehensive tutorial on Gaussian Mixture Models, featur‐
CGPA : 9.01 (as of 7th sem) ing visually engaging representations to enhance understanding of GMM
HIGH SCHOOL (2019‐21) with depth and clarity. Includes example visualizations in 1D, 2D, and 3D.
Green Valley High School 2022 xv6 Operating System Feature Addition C++, C
Percentage: 96% (Class 12) Added new features to MIT’s open‐source implementation of xv6 oper‐
ating system, like: System calls ‐ ’trace’, ’sigalarm’. Added scheduling al‐
gorithms like FCFS, LBS, PBS and MLFQ. Also implemented copy‐on‐write
fork.
2023 VLabs Web App JS, HTML/CSS, PWA, DynamoDB
Created a Web App for Virtual Labs IIIT‐H as a PWA (Progressive Web
Application), to cache the web page of each lab. Used AWS DynamoDB
as the database and deployed it to the Android Store with the help of the
Trusted Web Activities framework. Also designed the main homepage of
the app.
2023 Greddit React JS, HTML/CSS
A Reddit clone Web App using the MERN stack. Chatting website like
reddit, where users can add posts and comments, and get blocked or
reported. Users have different roles of admin, viewer or editor.
2022 Building an Interactive Shell C, Bash
Created a shell from scratch including basic commands like ’ls’, ’cd’, and
’cat’, and advanced bash functionalities like pipelining, signaling, fore‐
ground and background processes, and I/O redirection, using only the C
language.
2022 PID Control for Motor Angle C++, Arduino UNO
Project to control a motor adjusting both the motor power and direction
to gradually reach a specific angle over time by utilizing PID constants
and configuring the hardware components to showcase the application
of PID control.