Madhav Mahesh Kashyap

Implemented Django middleware to reduce master database load by rerouting SQL queries to alternate eventually consistent read-only replica databases. Resulted in a 40% reduction in the primary database’s CPU load.

Designed agent inline security upgrade pipeline to keep network-isolated systems up to date with the latest Debian package patches. My pipeline assisted with the Ubuntu 18.04 to 22.04 OS upgrade on 5,000+ running production agents.

Led development of configurable DNS feature enabling customers to set which public and internal DNS upstream servers their agents utilize. The dnsmasq based design worked on heterogeneous agent platforms including Virtualbox, Azure etc.

Conducted 8 technical interviews for Intern and SDE-1 positions.

Predicted systolic and diastolic blood pressures non-invasively for use in the ”Swasthya” cuffless health monitoring watch. Achieved 4.6% standard error using an 8th-degree polynomial regression model.

Developed an Angular2+SpringBoot full-stack web app to plot real-time patient biometric charts. Data was periodically polled from a FireStore database to update the frontend chart.

This paper tackles the problem of the heavy dependence of clean speech data required by deep learning based audio-denoising methods by showing that it is possible to train deep speech denoising networks using only noisy speech samples. Conventional wisdom dictates that in order to achieve good speech denoising performance, there is a requirement for a large quantity of both noisy speech samples and perfectly clean speech samples, resulting in a need for expensive audio recording equipment and extremely controlled soundproof recording studios. These requirements pose significant challenges in data collection, especially in economically disadvantaged regions and for low resource languages.

This work shows that speech denoising deep neural networks can be successfully trained utilizing only noisy training audio. Furthermore it is revealed that such training regimes achieve superior denoising performance over conventional training regimes utilizing clean training audio targets, in cases involving complex noise distributions and low Signal-to-Noise ratios (high noise environments). This is demonstrated through experiments studying the efficacy of our proposed approach over both real-world noises and synthetic noises using the 20 layered Deep Complex U-Net architecture.

PyTorch implementation open-sourced at GitHub with 100+ stars and 13 citations.

Despite unceasing debate about it’s pros and cons, exams and standardized testing have emerged as the main mode of evaluation and comparison in our increasingly competitive world. Inevitably, some examinees attempt to illegally gain an unfair advantage over other candidates by indulging in cheating and malpractice. Even a single case of examination malpractice can destroy an Examination body’s credibility and even lead to costly and time-consuming legal proceedings.

Our paper attempts to strategically allot examinees in specific seats and rooms, such as to mitigate the overall probability of malpractice. It involves examining multiple crucial factors such as subject similarity, distancing between examinees, and human field of vision to find the most optimal seating arrangement. We have exploited the property of Evolutionary Genetic Algorithms to find globally optimal or close to optimal solutions in an efficient time for this otherwise NP-complete permutation problem.