CS Ph.D. candidate at the New Jersey Institute of Technology (expected Aug. 2027), advised by Prof. Ioannis Koutis.
LLM inference/serving systems · learning-augmented algorithms · combinatorial optimization · graph and ranking methods · AI/ML systems
Python · C/C++ · ML tooling · optimization solvers (Gurobi, CPLEX, HiGHS, MiniZinc) · HPC/SLURM · reproducible research software
- llm-serving-heuristic-evolution — online scheduler portfolios for LLM-inference serving: GPU-calibrated simulation, trace-based counterfactual evaluation, contextual policy selection, and real-vLLM validation.
- consistency-aware-llm-rankin — consistency-aware ranking from pairwise preferences: cyclic preference-graph repair (MWFAS), retrieval evaluation, and statistical-inference methodology.
- Augmented-caching — decision-aligned eviction-value prediction for learning-augmented caching, with reproducible benchmark artifacts.
- combinatorial-opt-agent — retrieval-assisted instantiation of natural-language optimization problems: schema retrieval plus deterministic scalar-parameter grounding, validated on the NLP4LP benchmark.
- ranking-by-feedback-arc-set — training-free ranking from pairwise comparisons via acyclic graph construction (Journal of Supercomputing — accepted 2026).
- llm-serving-scheduler-robustness-benchmark — measures whether the best-ranked LLM-serving scheduling policy stays best across workload sources, operating regions, and evaluation metrics; reproducible benchmark archived on Zenodo and Hugging Face, with selected real-vLLM validation.
- Training-Free Ranking from Pairwise Comparisons via Acyclic Graph Construction — Journal of Supercomputing, accepted 2026
- Interweaving Real-Time Jobs with Energy Harvesting to Maximize Throughput — Algorithmica 2025
- Promoting Fairness and Priority in Selecting k-Winners Using IRV — KDD 2024
- Approximating Connected Maximum Cuts via Local Search — ESA 2023
- Parallel Longest Common Subsequence Analysis in Chapel — IEEE HPEC 2023