Danish Athar

dathar [at] purdue [dot] edu

danish8athar [at] gmail [dot] com

prof_pic.jpg

I am a direct Ph.D. student in Computer Engineering at Purdue University, where I conduct research as part of the Internet and Systems Lab (ISL@Purdue), advised by Prof. Sanjay G. Rao. Previously, I graduated with a B.S. in Computer Science (High Distinction, cGPA: 3.80/4.00) from Lahore University of Management Sciences (LUMS) in Pakistan.

My research interests lie broadly in AI systems and the intersection of computer networks and machine learning, particularly how each can be leveraged to improve the other. My current work focuses on distributed training of large language models (LLMs), including parallelism, scheduling, and training simulation, as well as datacenter networking and traffic engineering.

Previously, I was part of the Networks and Systems research group at LUMS, advised by Prof. Zafar Ayyub Qazi, Prof. Ihsan Ayyub Qazi, and Prof. Zartash Afzal Uzmi. There, I worked on LLM-based network measurement and automation, including router geolocation from reverse DNS hostnames, as well as semantic caching and web affordability.

I welcome research collaborations and internship opportunities in AI systems, networking, and distributed systems. Please feel free to reach out!

Education

  • Purdue University, West Lafayette, Indiana, USA
    Ph.D. Student, Computer Engineering
    August 2025 - Present
  • Lahore University of Management Sciences (LUMS), Lahore, Pakistan
    B.S. in Computer Science; cGPA: 3.8/4.0 (Graduated with High Distinction)
    Sep. 2021 - June 2025

Publications

  • Colt: Handling Traffic Bursts using Piecewise Linear Adaptive Traffic Engineering
    A. AlQiam, Danish Athar, N. Do, Y. Zhang, S. G. Rao, B. Ribeiro, M. Tawarmalani (Accepted to USENIX NSDI 2027)
  • Leveraging Large Language Models to Locate Routers: Using LLMs for Parsing rDNS Hostnames
    Danish Athar, M. Abdullah, K. Salamatian, F. Douzet, J. Cowie, Q. Lone, Z. Qazi, Z. Uzmi (Accepted to NewGeo @ ACM IMC 2026)
  • LLM-Enabled Semantic Caching for Affordable Web Access
    Hafsa Akbar, Danish Athar, Muhammad Ayain Fida, Hammad Javed, Dr. Zartash Afzal Uzmi, Dr. Ihsan Ayyub Qazi, Dr. Zafar Ayyub Qazi (Accepted to WiML @ NeurIPS 2025)

Research Experience

Selected research projects and coursework:
  • Distributed LLM Training and Simulation
    Research at ISL@Purdue on distributed LLM training using NVIDIA Megatron and simulation frameworks including SimAI and Sailor, with a focus on training parallelism, scheduling, and system performance.

  • Datacenter and WAN Traffic Engineering
    Research on traffic routing and performance using packet-level datacenter simulations in HTSim and analytical traffic engineering models in PyTorch.

  • Small Model Syndrome: Treatable with Prompting?
    Course Project for ECE57000: Artificial Intelligence. Study inspired by MedPrompt, done for much smaller model sizes. Shows that small generalist models can rival fine-tuned specialist models when guided by effective prompting. [Poster (BetterPoster format)][PDF]

  • L³R: Leveraging Language to Locate Routers!
    Undergraduate senior project at LUMS (PK) in collaboration with RIPE (NL) and GEODE (FR). Proposed a multi-agentic framework for IP geolocation and additional information retrieval using reverse DNS hostnames and large language models (LLMs). Evaluated the method on real-world hostname datasets from CAIDA used in previous geolocation research, matching and beating state-of-the-art techniques that require a lot more manual input. [PDF]

  • Embedding The Truth: Approximate Caching for Fact Checking
    Course Project for CS6303: Topics in Large Language Models. Explored the feasibility of reusing fact-checking claims across organizations and languages to improve efficiency and reduce costs. Used an approximate caching approach with multilingual embeddings in a vector database to identify recurring claims. [PDF]

  • Utilising LLMs for Streamlined Analysis of PTA Datasets
    Course project for the Topics in Internet Research course at LUMS. Idenitified ways for efficient analysis of Pakistan Telecommunication Authority (PTA) datasets using LLMs. [PDF]

  • A Framework for Improving Web Affordability and Inclusiveness
    Assisted the authors of this paper. Conducted a user study with 35 participants comparing web page quality and UX for Opera Mini, Brave and HBS-based affordable web frameworks. [Link to published paper (SIGCOMM 2023)]

Teaching Experience

  • Graduate Teaching Assistant, Purdue University
    • ECE36800: Data Structures (Fall 2025, Spring 2026, Fall 2026)
  • Head Teaching Assistant, LUMS
    • CS-582: Distributed Systems (Fall 2024)
    • CS-100: Introduction to Programming
  • Teaching Assistant, LUMS
    • CS-382: Network-Centric Computing (Spring 2025)
    • CS-382: Network-Centric Computing (Spring 2024)
    • CS-100: Introduction to Programming (Fall 2023 and Summer 2023)
  • Teaching Assistant, FutureTech
    • Served as the teaching assistant for the FutureTech (for more info on the Decoding The Internet module: a details site I made for my FutureTech students) sessions held at LUMS in summers of 2023 and 2024.
  • On-Site Instructor, CodeKids PK
    • Taught programming concepts to young learners using block-based and no-code platforms.

Work Experience

  • On-Site Instructor, CodeKids PK (June 2024 – July 2024)
    • Taught fundamental programming concepts to 20 young learners using block-based and no-code coding platforms, achieving a 95% satisfaction rate in post-class surveys.
  • Programming Intern, SenPi (Remote) (June 2022 – September 2022)
    • Worked on web development using the MERN stack.
  • Content Advisor, SenPi (January 2020 – December 2023)
    • Guided the development of high-quality educational content and instructional design, including strategic content planning, curriculum design, market research, and assessment design.
  • Content Development Specialist, OED Pakistan (Lahore) (October 2020 – October 2021)

Other Projects

  • Distributed Key-Value Store | Golang
    • Developed a fault-tolerant distributed key-value store using the Raft consensus algorithm, including leader election and log replication.
  • Distributed Hash Table | Python
    • Created a distributed hash table using socket programming, supporting a peer-to-peer model with failure tolerance.
  • Command Line Shell | C
    • Implemented a basic shell replicating UNIX shell functionalities, including pipelining and command chaining.
  • Basic File System | C
    • Implemented a simplified version of a UNIX file system that organizes memory into superblocks, inodes, and data blocks.
  • Second Time Around | MERN Stack
    • Developed a web application for a marketplace featuring auctions and donations with ML-powered recommendations.

Skills

  • Programming Languages: Python, C, C++, Go, JavaScript, TypeScript, Haskell
  • Research & Systems: Distributed LLM training, model parallelism (DP/TP/PP), network simulation, traffic engineering, network measurement, machine learning
  • Frameworks & Tools: PyTorch, NVIDIA Megatron, SimAI, Sailor, HTSim, Linux, Git, Docker, Apptainer, Gurobi, NumPy, Pandas, Selenium, Wireshark, Google Agent Development Kit (ADK), React, MongoDB, LaTeX

Awards

  • Graduation with High Distinction award at LUMS for outstanding academic performance during my undergraduate studies.
  • Dean's Honor List placement for all my years of study at LUMS.

Misc

  • In addition to problem-solving, I absolutely love cars and racing.
  • Returnal, a rogue-like game that counts on your failures for progress, is the hardest game I have completed, but has also been my most enjoyable gaming experience in recent times!
  • I also play a lot of League of Legends :D