Ayush Kumar

Experience

• 

  Project MANAS

  May 2021 - Jun 2023

  AI Subsystem Member

  Member of AI (Perception and Planning) department of Project MANAS: a student project team building India's first driverless car. - Worked on coverage path planning algorithms for an autonomous drone for AUVSI SUAS 2022- Developed a shape recognition model to classify aerial images of targets using invariant Hu moments

• 

  Chennai Mathematical Institute

  Feb 2022 - Jun 2022

  Research Intern

  - Working under the guidance of Prof. Mandayam Srivas on algorithms to prove permutation invariance in deep neural networks- Implemented efficient SVD algorithm for sparse matrices, ported solver code from Python to C++

• 

  CiSTUP (Centre for infrastructure, Sustainable Transportation and Urban Planning) @ IISc

  Dec 2022 - Feb 2023

  Research Intern

  - Developing mathematical models and transit route planning algorithms to predict congestion on Bangalore's BMTC bus transportation network- Parallelized route assignment pipeline using OpenMP and attained significant (68x) runtime speedup- Refactored data structures to optimise memory and runtime performance of multi-criteria route planning algorithm (McRAPTOR)- Developed mapping and visualization module to plot network-wide congestion over time

• 

  SAP

  Jun 2023 - Aug 2023

  Summer Intern

  - Built an interactive dashboard to streamline internal communications and event management workflows- Worked on a context-aware conversational agent to answer queries about employee benefits, along with a vector database for knowledge search

• 

  SAP

  Jan 2024 - Jul 2024

  Software Engineer Intern

  BTP SDK for iOS backend engineering team- Developed software tools for cloud application development on iOS and visionOS - Implemented support for a new development landscape (Edge), designed and implemented authentication flow using SAML for this landscape- Integrated SAP's Identity Authentication Service (IAS) with the SDK Assistant, allowing users to setup default authentication, single-sign on, and user management for their apps from an interactive wizard

• 

  Machine Learning Department at CMU

  Jan 2025 - now

  Research Assistant

  - Diffusion models in reinforcement learning, advisor: Prof. Max Simchowitz- Working on improving sample efficiency of diffusion methods in RL

• 

  Carnegie Mellon University

  Feb 2025 - now

  Collaborator @ Catalyst Research Group

  - Contributing to Mirage, a tool that generates fast GPU kernels for PyTorch programs through superoptimization techniques - Working on program partitioning framework to split and merge operators in a computational graph