Daniel Alexander
Computer Science PhD Student at Purdue University
Reconfigurable Networks · Distributed Systems · Hardware–Software Co-Design
I design algorithms and systems that coordinate application communication with dynamically reconfigurable network fabrics.
About
I am a Computer Science PhD student at Purdue University, advised by Prof. Vamsi Addanki in the STyGIANet group. My research interests lie in computer networks, distributed systems, and hardware–software co-design.
I study how application communication should be coordinated with dynamically reconfigurable network fabrics. My current work explores reconfigurable photonic interconnects, including the tradeoff between topology-reconfiguration overhead and multihop communication. More broadly, I am interested in scheduling, topology design, and workload-aware interconnects for emerging distributed systems.
Previously, I completed my MSE in Computer and Information Science at the University of Pennsylvania, where I worked with Prof. Vincent Liu on scheduling for LLM serving systems whose computation changes at runtime. I contributed to DREX, a dynamic rebatching system for early-exit inference, and developed ConFLExit, a confidence-guided scheduling mechanism, for my master’s thesis. Before Penn, I was a software engineer on ShopeePay’s core systems team at Shopee.
Current Research
Reconfigurable Photonic Interconnects
Reconfigurable photonic interconnects can change which nodes are directly connected, but every reconfiguration takes time. I am investigating when a network should reconfigure its connectivity to create more direct communication paths, and when the resulting reconfiguration delay outweighs the savings from avoiding multihop communication.
My current work develops simulation abstractions and workload-driven evaluations for studying this tradeoff, with longer-term interest in jointly optimizing workload placement, topology design, routing, and scheduling.
Selected Research
DREX — Dynamic Rebatching for Early-Exit LLM Inference
A serving system that dynamically reorganizes requests at each model exit point to improve batching efficiency, and handles the KV-cache state skipped layers leave behind with memory-efficient state copying. I contributed to scheduling and system implementation, experimental evaluation, and technical communication.
ConFLExit — Confidence-Guided Rebatching for Early-Exit LLM Serving
My master’s thesis on using model-derived confidence as a systems-level scheduling signal for router-aware flushing and dynamic rebatching. ConFLExit improves tail latency while exposing the tradeoff between responsiveness and batching efficiency.
Publications and Preprints
Dynamic Rebatching for Efficient Early-Exit Inference with DREX
Xuting Liu, Daniel Alexander, Siva Kesava Reddy Kakarla, Behnaz Arzani, Vincent Liu
Preprint, 2025 · Paper · Project
Application of Convolutional Neural Network for Semantic Segmentation of Bandung Urban Scenes
Daniel Alexander, Hans Christian Kurniawan, Irfin Afifudin, Hery Heryanto
IEEE International Conference on Data and Software Engineering (ICoDSE), 2022 · Paper · Project
Selected Systems Projects
PennCloud — Fault-Tolerant Distributed Cloud Platform
A distributed email and file-storage platform built on a custom partitioned key-value store. I implemented major parts of the replicated storage, coordination, failure detection, primary election, and recovery layers.
PennOS — UNIX-like Operating System
A UNIX-like operating-system project implemented in C, including preemptive priority scheduling, process lifecycle management, user-level concurrency, and a custom FAT-style filesystem.
Other projects: Mini Minecraft
Experience
Purdue University — STyGIANet Group
PhD Student, advised by Prof. Vamsi Addanki
2026–present
- Reconfigurable photonic interconnects: simulation abstractions and workload-driven evaluation of reconfiguration versus multihop communication.
University of Pennsylvania — Distributed Systems Lab
Graduate Researcher, advised by Prof. Vincent Liu
2024–2026
- Efficient early-exit LLM serving: dynamic rebatching (DREX) and confidence-guided scheduling (ConFLExit, master’s thesis).
Shopee — ShopeePay Core Team
Software Engineer
2022–2024
- Developed and maintained production backend services for ShopeePay’s core payment infrastructure.
- Worked in a reliability- and correctness-critical financial-services environment.
- Collaborated across engineering teams on service changes, incident resolution, and production operation.
Teaching
CIS 4190/5190: Applied Machine Learning, University of Pennsylvania
Head Teaching Assistant, 2025–2026 · Teaching Assistant, 2024–2025
Led approximately 15 teaching assistants for a course serving more than 200 students. Coordinated office hours, recitations, grading infrastructure, assessments, and large-scale machine-learning projects.
Education
Purdue University
PhD in Computer Science
2026–present
University of Pennsylvania
MSE in Computer and Information Science
2024–2026
Visiting undergraduate student, 2021
Harapan Bangsa Institute of Technology
Bachelor's in Informatics (Sarjana Informatika)
2018–2021
Selected Awards
- Ross Fellowship, Purdue University, 2026
- LPDP Scholarship, Indonesia Endowment Fund for Education, 2024
- Indonesian International Student Mobility Award, 2021
- Best Capstone Project, Google Bangkit Academy, 2020
Contact
Email: alexdan@purdue.edu · GitHub · LinkedIn
I am interested in research collaborations and internships in networked systems and emerging interconnects.