Hi,
I'am Advait Naik
Master's in CS student

Bennett University

Summer Research Intern

May 2021-Jul 2021

Headed a technical assessment to extract and generate skeleton data for Human Activity Recognition in drone-captured surveillance videos utilizing bottom-up 2-D Pose Estimation technique.

Formulated approach to integrated velocity and acceleration calculations with pose key points to provide comprehensive features for activity classification and constructed a Neural Network Architecture with an accuracy of 97.20% for categorizing activities as violent and non-violent.

Constructed Dynamic Graph Convolutional Networks for graph-based data analysis of Human Activity Recognition with skeleton data, achieving Top-5(%) model performance accuracy of 57.6% on drone-captured video dataset.

Computer Vision · R (Programming Language) · PyTorch · Statistics · Python (Programming Language) · Communication · Machine Learning · Artificial Intelligence (AI)

CERTIFICATE CLOSE WINDOW

Advanced Micro Devices, Inc. (AMD)

Software Engineer Intern

Jan 2024-Present

FastAPI · Azure · Docker · Python (Programming Language)

CLOSE WINDOW

USC Dornsife Center for Economic and Social Research

Software Engineer

Sep 2022-Dec 2023

Restructured Gateway to Global Aging Data - platform for contextual population survey data on aging from around world and tested infrastructure with production support (bug fixes, upgrades, refactoring).

Contributed to React development of microservices architecture, reducing the overall build time to almost 1/5th of the original build time.

Designed and Integrated highly scalable RESTful APIs for core microservices of Health & Retirement Studies for robust data management with SQLAlchemy and Python-Flask backend.

React · RESTful APIs · SQLAlchemy · Flask · Python (Programming Language)

CLOSE WINDOW

Leadingindia.ai

Machine Learning Intern

May 2020 - Jun 2020

Spearheaded team of 4 people to conceptualize an architecture for Violence Detection in drone surveillance with accuracy of 97.33% using transfer learning-based hybrid architecture of Convolutional Neural Networks (CNNs) and Recurrent Neural Network (LSTM)

Consolidated CNN model and Image processing techniques for Violent Individual Identification to overcome challenges posed by variations in human facial appearance in drone-captured videos and obtain 1.5x better result.

Leadership · PyTorch · Python (Programming Language) · Deep Learning · Machine Learning · Artificial Intelligence (AI)

CERTIFICATE CLOSE WINDOW

VESIT Renaissance Cell

Web Developer

May 2020 - Jun 2020

Collaborated closely with a team of 6 members and designed a highly interactive and responsive website for the College Canteen Management system (E-Canteen) using React and NodeJS frameworks.

Implemented MongoDB database to efficiently managed a large volume of data, storing and processing >2000 orders on the platform, and leveraged its efficient indexing and querying capabilities to suggest similar items to customers based on their preferences and order history, resulting in increased customer satisfaction.

The website achieved a 2x better speedup in terms when benchmarked against the existing website, making it faster and more efficient.

Facilitated team coordination and headed project engagement tasks of communicating project updates; documented technical specifications of application architecture and workflow.

The project provided valuable experience in web development, data management, and collaboration with a team, and we were able to deliver a successful project that met the client's requirements.

MongoDB · Agile Methodologies · React · Node.js · HTML · JavaScript · Git

CERTIFICATE CLOSE WINDOW

KIDsVIEW - A PARENTS COMPANION

Pradarshan'21

(CMPN Technology Day)

First Prize

Increase in the employment rate of mothers with children under 6 years old. Parents forced to leave their child at home with a caretaker. Effect on the growth and development of the child. Our system helps the parent track their children’s activities and also gives the child's parent an emotional analysis. Safety of the child with the caretaker is also a concern.

Introduced benchmark datasets for child activity recognition and child abuse detection with over 1200 seconds of 4 activities and abuse-non abuse video clips. Performed skeleton-based child activity recognizing with an accuracy of 91.27% deploying Random Forest model and depth-wise Separable Convolution Neural Network for child emotion recognition. Monitored and managed deployment of AbuseNet model for detecting child abuse attaining 98.20% accuracy

VIDEO LINK CODE LINK CLOSE WINDOW

RESTAURANT WEBSITE

Build a software restaurant automation website that will have several features-

Stock management

Internet online reservation

Online ordering via network enabled handheld computer

Customer management

Accounting

Executive information system

VIDEO LINK CODE LINK CLOSE WINDOW

AUTOMATIC ATTENDANCE SYSTEM

National Innovation Contest 20

Institution's Innovation Council

In India, the reopening of educational institutes is delayed due to the fear that students may not follow basic corona precautionary norms. So we aim to create an AI-based system to check whether students are following all rules or not and scan the body temperature, for precautionary measures. We aim to create an automated attendance system that will promote contactless and hassle-free attendance. Our system would reduce the fear of reopening educational institutions by making monitoring of students easy and would make the students feel safer given the fact that they are being monitored with all the necessary precautions.

VIDEO LINK CODE LINK CLOSE WINDOW

METAL DETECTOR

Awakening the Scientist 2018-19

Third Prize

The metal detector works on an oscillator producing an alternating current which produces a magnetic field in a coil leading to induction of eddy currents in the metals

CERTIFICATE CLOSE WINDOW

Multimodal Detection and Analysis of Parkinson Disease

BE Project

Computed Composite Feature Score (CFS) of motor movements, sketching, pen pressure, and vocal impairments by leveraging diverse multi-feature processing techniques to reveal how an automatic detection technology can support a neurologist to develop accurate scores up to 92.7% for evaluating Parkinson's Disease (PD) decision-support systems

VIDEO LINK CODE LINK CLOSE WINDOW

Yelp Web App To Locate Businesses

Web Technologies

Developed a multimodal application for discovering local businesses ranging from bars, restaurants, and cafes to hairdressers, spas, and gas stations using 6+ APIs and deployed using Google Cloud App Engine.

Designed 5+ application features for pages devoted to individual locations, such as restaurants, including assisted business search, automated geo-location, crowd-sourced reviews about businesses, and in-app reservation functionality and provided a one-stop local platform for consumers to discover, connect and transact with local businesses of all sizes that develops the website and the mobile app.

VIDEO & CODE LINK CLOSE WINDOW

Evolutionary Algorithm Neural Architecture Search for CrisisMMD Dataset

Natural Language Processing

University of Southern California

The project proposes an evolutionary algorithm for neural architecture search in natural language processing (NLP) tasks. The proposed methodology involves exploring the dataset, preprocessing the data, representing neural networks as a genome, and implementing the evolutionary algorithm. The neural networks are represented as directed acyclic graphs (DAGs), and the evolutionary algorithm generates an initial population of DAGs, applies genetic operators - crossover and mutation, and evaluates the fitness of the resulting DAGs using the CrisisMMD dataset. The proposed method aims to construct formidable neural networks for NLP tasks, providing a useful technique for classification tasks. These results are compared to traditional Machine Learning methods.

CODE LINK CLOSE WINDOW