Education

Research Experience

Pre-Doctoral Research Fellow

Microsoft Research India (MSRI) | July 2024 – Present

Mentors: Dr. Debopam Bhattacherjee, Dr. Rohan Gandhi, Dr. Anjaly Prayil

Projects:

• Output Length Prediction for LLMs
  • Designed predictors for output sequence length using query context.
  • Enabled more efficient routing, scheduling, and memory management in LLM serving pipelines.
  • Evaluated predictors across multiple LLMs to ensure generalizability.
• Dynamic Response Length Control
  • Built a system that dynamically adjusts response lengths under high GPU utilization.
  • Reduced energy consumption while preserving information fidelity.
  • Verified approach through large-scale experiments with real workloads.

Undergraduate Research Assistant

Indian Institute of Technology Dharwad (IITDh) | April 2022 – May 2024

Affiliated to Future Generation Lab

Mentor: Dr. Koteswararao Kondepu

Projects:

• Energy Analysis of different 5G RAN Architectures
  • Assisted in deploying ORAN architectures: Monolithic, Dis-aggregated, and CUPS.
  • Investigated energy consumption for RAN and UE using S-tui and Scaphandre.
  • Found a 19.3% energy increase when UEs scaled from 1 to 10 in disaggregated RAN.
• Resource Allocator
  • Developed a MERN stack application for remote resource allocation using JWT sessions.
  • Used Docker to create containers with specified resources for users.
  • Built an open-source FPGA toolchain integrated into the platform.
• Edge Assisted Autonomous Surveillance using UAVs
  • Annotated 2000+ images to prepare a custom dataset for object detection.
  • Trained YOLOv3 & YOLOv5 models on custom dataset and containerized them.
  • Developed scripts to control UAVs based on model inferences.
• Automated Controlling Mechanism
  • Implemented Grafana + Kafka framework to alert on CPU and memory over-utilization.
  • Automated Slack alerts at 70% utilization; auto-intervention at 80% usage.

Mitacs Globalink Research Intern

Lakehead University, Ontario, Canada | May 2023 – July 2023

Mentor: Dr. Shafiqul Hai

Projects:

• Hardware-Software Codesign of a Convolutional Neural Network
  • Designed CNN for MNIST digits with 98% accuracy; optimized for 1M pixel images in 20ms.
  • Implemented block circulant matrix + 16-point FFT to reduce FPGA logical elements by 40%, maintaining 96.89% accuracy.
  • Co-authored journal paper: "Power Efficient CNN using Pipelined FFT Architecture" accepted at Embedded Systems Letters.