Projects
Maximizing Datacenter Power Efficiency through Local Load Imbalancing
Master’s thesis, Georgia Tech | Advisor: Prof. Ahmed Saeed | 2025–2026
- Designed the Slack-Aware Load Imbalancing Protocol to concentrate latency-sensitive traffic on fewer servers, allowing unused capacity to stay in low-power idle states.
- Built a distributed proxy, coordinator, and load generator in Go and gRPC and evaluated Memcached and NVIDIA Triton workloads on a 10-node bare-metal CloudLab cluster.
- Measured up to 7% cluster-wide power savings (14% in simulation) while meeting tail-latency targets. Linux networking and IRQ tuning increased single-node throughput capacity by 26%.
Video Transmission with Fountain Codes
Mentor: Prof. Jun Xu | Sep–Dec 2024
[Report] [Presentation] [GitHub]
- Benchmarked LT and Raptor erasure codes in Python, measuring decoding overhead, recovery, and scalability under different channel conditions.
- Built an H.264 and FFmpeg pipeline that applies erasure coding based on scene changes, reducing transmission time under high packet loss while keeping video quality close to the baseline.
Generative AI Discord Bot
July 2024
- Created a Discord bot that integrates with Google’s Generative AI and Wit.ai to provide text and voice-based functionalities.
- Implemented features like text generation, user attitude summarization, voice channel interaction, and voice transcription using Wit.ai.
NVIDIA BlueField DPU and Kernel-Bypass Characterization
Mentor: Prof. Sameer G. Kulkarni | Aug 2023 - Dec 2023
- Conducted a study on NVIDIA's BlueField DPUs, characterizing three operational modes and evaluating four logical modes for latency, throughput, and application benchmarks.
- Compared the kernel IP stack and kernel-bypass paths with Nginx, Apache, Twemproxy, and Memtier workloads to identify where hardware offload improves latency and throughput.
ZARTH Augmented Reality Astronomy App
Research intern, Caltech | Mentor: Dr. Ashish Mahabal | May–Jul 2023
- Developed Android and backend features for a citizen-science app that lets users discover astronomical transient events.
- Used Firebase and Firestore for accounts, leaderboards, and real-time data synchronization. Optimized Python filters for selecting transient events from large datasets.
On The Scalability of Temporal Graph Neural Networks
Mentor: Prof. Anirban Dasgupta | Feb 2023 - Apr 2023
- Investigated the scalability of Temporal Graph Neural Networks (GNNs) in dynamic graph representation learning.
- Built a graph dataset of daily rainfall in the Indian subcontinent from 1900 to 2018 and compared GraphSAGE and Roland for scalability and catastrophic forgetting.
Gossip: Python-based Interpreter
Mentor: Prof. Balagopal Komarath | Jan 2023 - Apr 2023
- Developed a Python-based interpreter/compiler for Gossip-lang, supporting concise coding for small projects.
- Implemented features such as static type checking, loops, and exception handling.
Comparative Analysis of Data Plane Packet Processing Modes
Mentor: Prof. Sameer Kulkarni | Aug 2022 - Dec 2022
- Conducted a comparative analysis of traditional kernel mode versus kernel bypass mode (DPDK) for data plane packet processing.
- Evaluated factors such as throughput, latency, and resource utilization, demonstrating the performance benefits of kernel bypass mode.
Chip8 Emulator in Python
December 2022
- Developed a simple Chip8 emulator in Python based on the tutorial at Multigesture.
- The emulator supports basic Chip8 functionality, allowing users to pass a ROM file and run it in Python.
- Controls for the emulator are mapped according to Chip8 specifications.
Web Cookie Analyzer
October - November 2022
- Developed a tool for scraping and analyzing web cookies of a given URL using the Selenium library in Python.
- Implemented functionality for retrieving the TLS version of a site using
tsharkCLI and subprocess API.
Remote File System using Python
September 2022
- A simple implementation of a remote file system using socket programming in Python supporting various
file
functions like
ls,cwd,dwd,upd, andcd. - Implemented using different networking layers: encryption, network layers, and application layer.
Statistical Language Modeling using N-Grams
Mentor: Prof. Mayank Singh | Feb 2022 - Apr 2022
- Developed a trigram statistical language model for next-word prediction, deployed on HuggingFace.
- Analyzed perplexity scores on English Wikipedia articles using Markov assumptions and Laplace Smoothing.
Jacobian Transpose Method for Controlling Robotic Manipulators
April 2022
- Performed path tracking on a two-arm robotic manipulator using the Jacobian transpose method in MATLAB.
- Simulated this algorithm in Unity for real-world applications of robotic control.
MIPS Pipeline Simulator
March 2022
- Designed a 5-stage discrete event MIPS simulator using Python and Tkinter library.
- Implemented hazard detection and applied techniques like data forwarding to mitigate pipeline hazards.
Reinforcement Learning using Unity's ML-Agents Library
September 2021
- Implemented reinforcement learning using Unity's ML-Agents library for a machine learning application.
Modeling Gene Expression using Differential Equations
November 2021
- Modeled the Gene Expression process by constructing algorithms for transcription and translation using differential equations.
- Solved the equations linearly and visualized the results using MATLAB.
Publications and Posters
-
When Server-Level Load Imbalance Helps: Power Savings without Capacity Loss
Sherif Abdelrazek, Dhyey Thummar, and Ahmed Saeed. USENIX NSDI poster, 2026. [Poster PDF] -
Distributed In-band Network Telemetry
Dhyey Thummar, Iram Nawab, and Sameer G. Kulkarni. IEEE/ACM CCGrid Workshops, 2023.