Selected work · Archive
A collection of open-source, research, and earlier projects beyond the featured case studies.
University of California, Los Angeles · Jun 2022 – Feb 2023
At UCLA, developed a text-diversity framework and quality predictor, alongside an encoder-decoder model that generates paraphrases with targeted syntactic, phonological, and morphological diversity while preserving meaning. Trained models were published on Hugging Face.
Google · Jun 2022 – Sep 2022
Improved Ad Load coverage by 8% for queries without historical interaction results by extrapolating from similar queries.
University of California, Los Angeles · Apr 2022 – Jun 2022
Compared network-compression methods on established deep-learning models such as ResNet using accuracy, inference time, and model size, and presented the analysis and generation strategies using CIFAR-10. Also analyzed combined model performance and ensembling, including inference gains through core-kernel analysis on a Titan V GPU.
Standalone methods reduced model size by up to 40× with under 5% accuracy loss. Ensembling improved accuracy by a further 2.5% at the same combined model size. Combining compression methods produced a 215× smaller model with only an additional 1% accuracy loss and up to 8× lower inference latency.
University of California, Los Angeles · Jan 2022 – Mar 2022
Performed multi-class classification on EEG data from the BCI Competition and compared CNN, CRNN, and autoencoder architectures across experiments, including single-subject versus all-subject training and different time durations.
University of California, Los Angeles · Jan 2022 – Mar 2022
Created a Python quantum-circuit simulator that compiles and runs QASM programs. Implemented algorithms including Shor’s, Grover’s, and Bernstein–Vazirani using Cirq and Qiskit.
Microsoft · Jun 2019 – Aug 2019
Built a volunteering app where NGOs and small groups could post activities, volunteers could apply to activities matching their preferences, and matching volunteers received notifications.
National Institute of Technology Tiruchirappalli · Jul 2018 – May 2019
Developed a Conditional Random Field financial named-entity recognizer that achieved 92% accuracy in identifying financial key phrases. Upgraded question-generation and answer-extraction modules to use the extracted entities, increasing answer-extraction ROUGE-L from 0.909 to 0.969.
Contributors: Hariharan and Vishal
Read the Sādhanā paper ↗ View the project repository ↗Microsoft · May 2018 – Jul 2018
Reduced page-load latency by 6× and application boot time by 42× using caching and authentication-flow improvements. Added web-parity, System Tray, and background-task capabilities while separating the desktop app from its WebView-hosted app.
National Institute of Technology Tiruchirappalli · Apr 2018 – Present
Created an Android app for sharing a Wi-Fi network by scanning a QR code rather than sharing its password. Used ZXing for QR-code generation and added fingerprint authentication.
National Institute of Technology Tiruchirappalli · Jan 2018 – Mar 2018
Built an image-captioning project using a convolutional neural network and a recurrent neural network to generate a sentence describing an image, then convert it to speech. The project aimed to help visually challenged people follow their surroundings through a live relay of events.
Contributor: CR Anish
View the project ↗IIT Madras · May – July 2017
Studied deterministic dispersion of mobile robots on a ring under vertex permutation and 1-interval connectivity. The work was presented at ICDCN 2018.
Read the paper ↗National Institute of Technology Tiruchirappalli · Jul 2016 – Present
Built a web application for retrieving movie information from the OMDb JSON API. AJAX requests send data to the web service and update the application dynamically.