Working on using Diffusion Models for Multi-Agent Planning in Robotics Systems.Building an end-to-end system in Jax. Under Prof. Suresh Jagannathan

Open Source AI: AutoGluon (https://auto.gluon.ai/)Built an Automated RAG (Retrieval Augmented Generation) package from scratch. https://github.com/autogluon/autogluon-ragUsers can obtain a production-level, inference-ready RAG pipeline for their documents in three lines of code. Package Supports multiple models, vector databases, reranker, and retrieval methods. Each module can be customized individually from high-level models to low-level API parameters.

Research Assistant working on Efficient Path Planning, Reinforcement, and Verification in Robotics using LLMs. Submitted paper to ICRA 2025.Under Prof. Suresh JagannathanAssociated with: Prof. Lin Tan, and Prof. Aniket Bera

AWS Deep Engine-Science - Open Source AI: AutoGluon teamAutoGluon is an open-source AutoML library that automates machine learning and deep learning tasks to easily achieve strong predictive performance in user applications. Users can train and deploy high-accuracy deep learning models on image, text, time series, and tabular data in just a few lines of code. AutoGluon utilizes modern deep-learning techniques like multi-layer stack ensembles to fit the user’s needs.Worked on model benchmarking, evaluation, and dashboards to track the performance of AutoGluon across different datasets, frameworks, and model hyperparameter configurations. Also worked on dashboards for visualizing benchmark metrics from AWS EC2 instances. This tool is used to analyze the effects of a change made to the AutoGluon framework. Additionally, created a website to compare various AutoML frameworks across several datasets and configurations.Autogluon Website: https://auto.gluon.ai/stable/index.htmlAutoGluon Dashboard: https://github.com/autogluon/autogluon-dashboardAutoGluon Benchmark: https://github.com/autogluon/autogluon-bench

Co-Head Lab TA for EECS 16A, Designing Information Devices and Systems I. (Average Class Size: 1000 students)Led a lab staff of 20-30 students to lay out lab syllabus and curate productive lab experiences for students. Assigned to teach students the process of designing and debugging circuits as well as writing iPython code in Jupyter notebooks for machine learning and signal processing. Projects done in the semester: Single-Pixel and Multi-Pixel Imaging, Resistive and Capacitive Touchscreen models, Acoustic Positioning System.Also assigned to write and debug exam problems for midterms and finals as well as grade papers. Core concepts covered in EECS 16A are signal processing, machine learning, control, and circuit design while introducing key linear-algebraic concepts motivated by real-world applications. Labs allow students to engage computationally, physically, and visually with the concepts they learn in lecture.

Worked under the Amazon Operations org in the Telemetry team. Overarching project was a Self Healing network, where the network running at Amazon Fulfillment Centers heals itself when a software/hardware issue arises. Created an API to automate, previously manually run, searches made by network engineers to remediate network issues outlined on auto-cut tickets.

-Overseeing team project collaborations with various prominent Neurotech startups/companies-Supervising project development: •From a technical perspective, providing development frameworks and technical guidance to teams and project managers. •From a leadership perspective, guiding the project managers in building strong teams and creating an innovative and collaborative work environment. -Growing the power and brand of NTAB: •Working with the outreach division to put on industry events, by connecting with several industry leaders

Neurotech@Berkeley's software division is committed to using EEG data to build innovative, impactful software solutions to neuroscience problems. Most of our projects involve writing machine learning algorithms using the data that we collect from the EEG headsets. Previously worked on using signal processing and ML techniques to filter and classify EEG/EMG data in order to control a small electric car using the brain.Libraries: scipy, numpy, matplotlib, pandas, seaborn, mne

NeuroTech@Berkeley is a student-run neuroscience club that focuses on using non-invasive techniques such as EEG (Electroencephalography) for cognitive research and development. As a Publications member, I wrote various research based articles for NT@B covering topics on the significance of ongoing research work backed with scientific reasoning. Some of my articles include work on neuroprosthetics, AI, and Virtual Reality. My articles have been featured on medium as well as published on a couple of NeuroTech magazines. https://medium.com/neurotech-berkeley

HPCaaS: Developing and deploying High Performance Computing (HPC) microservices through workload-optimised systems as a part of HPE’s umbrella cloud service, commonly known as Greenlake. Working to provide users with a platform to perform tasks on HPE’s supercomputer clusters.1. Worked on a Backup and Restore procedure in case of catastrophic failure of supercomputer clusters. Created a cron job that constantly takes a backup of the MySQL database and moves it to a Backup Directory for future manual restoration of the supercomputer cluster and states of all computing nodes. 2. Created a cache package in GoLang to cache the JWT tokens in the Slurm (job scheduler for Linux and Unix-like kernels used for supercomputer clusters) endpoint API. Reduced the response time for the API by 66.67% due to the release of this cache package. 3. Incorporated an ELK Stack (Elasticsearch, Logstash, Kibana) (Engine for data analytics, Server‑side data processing pipeline and Powerful visualizations) to view analytics for jobs on the supercomputer clusters. Created a dashboard on Kibana to view fields like No. of Jobs Submitted, Wait Times, Failure and Success Rates, etc. Other Tools worked with: Vagrant (Tool for building and managing VMs), Ansible API (Automation engine for cloud deployment, management and resource provisioning)

• Leading a team of 5 undergrads from UC Berkeley (members of Neurotech@Berkeley) to develop two software tools (web apps) to be used in Neurable’s public demos of their latest, cutting-edge Brain Computer Interface (BCI) headset. • Developing real-time streaming and visualization infrastructure for EEG data obtained from the user’s BCI headset.• Developing an electrode heatmap which changes colors (from Red to Green) based off calculated impedance (Z) threshold values. This heatmap can be used by the user to configure Neurable's EEG headset. • Developing a neuroscience-based experiment for the users to undergo, to exhibit the efficacy and performance of Neurable’s headset. • Tools used: JavaScript, Flask, Python, plotly, LSL streaming packages, Data analysis libraries (pandas), SciPy

• Headed the development of a real-time EEG neurofeedback game using Unity3D. • Included an EEG streaming infrastructure for real-time ML processing and visualizations of raw data, spectrograms and various cognitive features.• This project was done by a team of 6 undergraduate students from NeuroTech@Berkeley, a neuroscience club at University of California, Berkeley.

Worked with the Emerging Technologies department (DevOps) at Injazat Data Systems, Abu Dhabi. Introduced to projects in Artificial Intelligence and Augmented Reality.AI: Healthcare sector - Product that applies deep learning algorithms to medical imaging data such as X-rays, CT and MRI scans to be used to assist doctors in a broad range of image diagnosis, illness recognition and classification tasks.AR: An extended reality visualization tool that uses advanced computer vision, calibrated headsets (HoloLens/VR Goggles), and cross-platform reach (mobile version-iOS and Android) to allow industrial engineers to develop and execute infrastructural plans in a 3-D holographic model. Additionally, field engineers can use the headsets to see what is beneath their construction sites (X-Ray like view) to simplify the location of utilities and other underground infrastructure.

Learnt the foundations of autonomous driving, important algorithms and concepts related to the creation of an autonomous vehicle. Topics covered were: System ID, Braking, Convolutions, Deep Computer Vision, Path Planning (Dijkstra's Algo, A* Search) and Control Theory. Created the basic elements of a self-driving car in a simulated virtual environment that had a pre-designed track, stop signs, pedestrians, signals, etc. Worked on the data collected from the simulations to improve the self driving model. Tools used: Python, numpy, tensorflow, pytorch, nn

Head of Event Management for ClubRiseAD, a student run socio- environ club that was formed with a sole motive to spread awareness about critical issues that prevail in our society. Organized fundraisers, public-area cleanups, anti-smoking/anti-vehicle idling campaigns, flood relief donation camps, winter clothes donation, etc. 100+ members currently in the club.

Core member of the organizing team of TEDxYouth@ADIS platform during 2017 and 2018. Also, one among five speakers in the inaugural TEDxYouth program held in 2017.* ADIS = Abu Dhabi Indian School (my former alma mater)

Volunteered at Cleveland Clinic Abu Dhabi. Worked with various functional departments in the Hospital such as Emergency Department, Cardiology, and the Robotics division used to train and test doctors. The robots reacted to various external stimulants (touch, temperature, pressure)