Mercedes-Benz Research and Development

Research Engineer

Present

I am working on developing novel V2P commmunication architectures between Mercedes vehicles and mobile application. We are currently inetgrating the Cat-M1 and MQTTv5 technologies into our proposed plan of action. As part of this development, I have been researching on latest technolgies, and current architecture style used around. As this is an end-to-end task, we are also looking at the deployment of this technology using cloud-based microservices, primarily hosted over Azure and developed using Java (Spring). Lastly, as part of the next-gen Mercedes vehicles, I am working on the development of a low-cost, P2P communication method between the vehicle and the Mercedes-Me app (on mobile phone) via BLE and UWB.

SwiftNav: A probabilistic global optimizer

Prof. Debasish Chatterjee | Department of Systems and Controls Engineering, IIT Bombay

Created a novel global optimization algorithm (SwiftNav) based on probabilistic Markov-chain Monte Carlo (MCMC) samplers. It aims at obtaining the global minima for a given function over a compact domain. We utilized the Walker-Slice sampler and implemented Gibbs sampling for robust high-dimensional computations. Further, the search space for any function over the compact domain was discretized dynamically -- meaning that the grid could be refined or made coarser on the fly. Lastly, parallel processing was employed to make high-dimensional optimizations quick.

Algorithmic construction of Controlled Lyapunov functions (CLF)

Prof. Debasish Chatterjee | Department of Systems and Controls Engineering, IIT Bombay

We utilized a computationally tractable algorithm (MSA Algorithm), coupled with SwiftNav which allowed us to algorithmically construct CLFs. CLFs are an important centerpiece in control engineering and in the field of Machine Learning. It can be used for multiple use-cases spanning across a diverse field like Spacecraft attitude control, Epidemic mitigation, Catalyst control, and even for biochemical reactions. We have examined these applications in detail. A bunch of polynomial and trigonometric functions were used to define the posible CLFs over the discrete space and the results were validated against the constraints specified.

DrugProtAI: Protein druggability predictor

Prof. Sanjeeva Srivastava | Department of Biosciences and Bioengineering, IIT Bombay

We developed a comprehensive tool used to evaluate the druggability potential of 20000+ human proteins. We introduced a novel Partition Ensemble Classifier (PEC), running multiple ML models that analyse 183 biophysical and protein sequence-specific properties. Our website acts like a knowledgebase, offering druggabilty predictions and access to 2M+ publications on drug targets along with multiple additional features.

CovProt: Mass spectrometry based pretomics data resource

Prof. Sanjeeva Srivastava | Department of Biosciences and Bioengineering, IIT Bombay

I was a part of the team that developed a comprehensive resourcebank hostig proteomics data of SARS-CoV-2 infected patients to facilitate COVID-19 research. We conducted research on the proteomics data in various organs of COVID infected patients and studied their variation. The website aims at providing substantial coverage of host-related proteomics data of SARS-CoV-2.

Optimal sphere packing in higher dimensions

Prof. Avinash Bhardwaj | Department of Industrial and Operations Research, IIT Bombay

As part of my BTP, I Conducted research on mathematical models formulating spheres as d-dimensional lattices to maximize packing density. We examined the Monte-Carlo approach, along with the TJ and LS algorithms. This problem is of interest in cryptosystems and finds multiple applications in computational mathematics.