MS Computer Engineering @ NYU. I work on ML systems: distributed training benchmarks, evaluation tooling, and reproducible experimentation.
- Distributed Training & Benchmarking: fixed-work experiments, throughput/step-time measurement, multi-GPU scaling (PyTorch + DeepSpeed/ZeRO)
- Systems Foundations: C++ projects (cycle-accurate simulators), Linux tooling, test harnesses
- Open Source: Contributed to Opik (Comet ML) Python SDK — merged PR with unit tests + docs
| Project | Tech Stack | Evidence / Impact |
|---|---|---|
| Distributed LLM Training Benchmarks | PyTorch, DeepSpeed, Slurm | Fixed-work multi-GPU benchmarking for GPT-2 (124M); tokens/sec scaling and run artifacts for reproducibility. |
| MIPS Processor Simulator | C++, Linux, Make | Cycle-accurate 5-stage pipeline simulator with hazard detection + forwarding; verified via regression tests / traces. |
| Opik LLM Eval Platform (Merged PR #1006) | Python, Pytest, Docs | Added SentenceBLEU/CorpusBLEU metrics (NLTK-backed) + unit tests + docs; exported via opik.evaluation.metrics. |
| Brain Tumor Segmentation Baseline (MONAI 3D U-Net) | PyTorch, MONAI, Linux, Slurm | Reproducible training/eval pipeline on MSD Task01 with guardrails (ROI/label checks, metric conventions) and saved artifacts for reruns/plots. |
- Systems: Python, C++, Linux, Bash
- ML Systems: PyTorch, DeepSpeed (ZeRO), testing (pytest), experiment reproducibility (configs/artifacts)
- Tooling: Git, Docker
