Papers and detours
Write-ups and numbers included. GitHub links sold separately.
“Beyond Utilization: Energy-Conscious GPU Sharing for Inference Serving”, 3rd author, SOSP '26 (published). EnerTune, from Yadwadkar's SysML Lab at UT Austin. I rebuilt the four GPU-sharing serving systems it is measured against and ported them to our 16-A100 cluster, enabling its 1.4-2.3x lower-energy result, and contributed baseline profiling to its 7.3x average profiling-time reduction.Read paper (ACM) ↗EnerTune repo ↗The baselines I built →ORCID ↗
Featured
current research & systems- 01
EnerTune Baselines
Four published GPU-sharing serving systems rebuilt from their papers and ported to a 16-A100 cluster to serve as EnerTune's baselines.
- 02
MLLM-Energy
1,080 profiled configurations of multimodal LLM serving on an A100, measuring how much of the GPU each inference phase uses and what it costs in energy.
- 03
CHT-Radix
Seven concurrent hash tables and radix trees in C++, benchmarked against both synthetic traffic and a real ShareGPT prefix-cache trace.
- 04
Pintos
The Pintos teaching kernel built out in C: user processes, system calls, demand-paged virtual memory, and full file system support.
Everything else
2022–2025- 05YASHA Unix shell written from scratch in C: command parsing, pipes, redirection, and job control.
- 06GoodEatsCommunity platform for local food deals, built in the SEO Tech fellowship and used by 50+ UT students.
- 07EcoLensA pantry scanner that recognizes food items and scores their environmental impact.
- 08PS-AmperesA custom current-sensing board designed for the UT Longhorn Racing electric vehicle's battery system.
- 09Primate-SimAn instruction-accurate RISC-V simulator built in Chiou's FAST Lab at UT Austin to validate the Primate compiler.
- 10Astro Party EmbeddedAn Astro Party clone for ECE 319K, with real-time gameplay running bare-metal on a microcontroller.
- 11Keyboard2x2My first PCB: a 2x2 mechanical macro pad designed in KiCAD to learn hardware design.
- 12Variational Quantum EigensolverA high school team project implementing a quantum algorithm for molecular ground-state energies.