Resume

Current resume of Aman Pandey Sanjeev Kumar: LLM inference, GPU kernel engineering, compiler performance analysis, and edge AI deployment.
LLM Inference & GPU Performance
Download ResumeOpen PDFView ProjectsGitHubLinkedIn
Current resume of Aman Pandey Sanjeev Kumar. Project summaries, skills, education, and experience follow below.

Technical Focus

GPU Kernel Engineering FP32 RMSNorm and GEMV/small-M GEMM in CUDA C++ and Triton; fused seven eager launches into one and passed 80/80 backend-shape checks against PyTorch

Single-GPU LLM Runtime Continuous batching, chunked prefill, paged KV tables, and prefix caching; 58 → 223 tokens/s versus single-sequence scheduling on 12 mixed-length Qwen3-0.6B requests

Compiler Performance Analysis FX capture, torch.cond, isolated caches, and fresh-process profiling; Qwen2.5-0.5B FP32 eight-step decode 97.9 → 66.8 ms with CUDA Graphs disabled

Triton Forward Attention FA2-style online softmax, FP16 MHA/GQA, causal tile pruning, and partial-tile masking; 60/60 CUDA forward cases validated against PyTorch SDPA

Edge AI Deployment MobileCLIP2-S2 via PyTorch → ONNX → Qualcomm QNN; image/text P90 14.51/3.99 ms on S22 Family, with 794/794 profiled operations on NPU

Technical Skills

  • Programming & GPU kernels: Python, C++, CUDA, Triton, CuTe DSL, CUTLASS.
  • Frameworks & compilers: PyTorch, torch.compile, TorchDynamo, TorchInductor, FX, ONNX, MLIR, XLA, TVM.
  • LLM inference & serving: vLLM, SGLang, TensorRT-LLM, continuous batching, KV and prefix caching, prefill/decode scheduling, TTFT, TPOT/ITL, throughput, goodput/SLO analysis, and tensor/pipeline/expert parallelism.
  • Profiling & benchmarking: PyTorch Profiler, Nsight Systems, Nsight Compute, Perfetto, benchmark harnesses, warmup/steady-state timing, and correctness validation.
  • Quantization & precision: PTQ, AWQ, SmoothQuant, scaling granularity, symmetric/asymmetric integer quantization, INT8/INT4, FP16/BF16, FP8, NVFP4, NF4.
  • Systems & deployment: Linux, Docker, Kubernetes, Slurm, Git, Bash.

Education

Purdue University · West Lafayette, Indiana

Master of Science in Electrical and Computer Engineering · May 2026

Specialization: Artificial Intelligence & Machine Learning

Experience

Teaching Assistant, IE 332: Computing in Industrial Engineering

Purdue University · August 2023 to December 2025

  • Led labs for 75 students on machine learning algorithms, data pipelines, and supervised/unsupervised learning.
  • Taught relational database design, ERDs, SQL, and backend web development with HTML/CSS/JS/PHP/MySQL.
  • Guided student projects integrating ML with ERP, CRM, and SCM systems.

Data Science Researcher

Cisco × Purdue Data Mine · January to May 2024

  • Built forecast selection and ensembling algorithms for 10,000+ Cisco products, improving cost-weighted prediction accuracy by 3%.
  • Clustered PLIDs with Gaussian Mixture Models to refine forecast segmentation.
  • Applied Bayesian MCMC, MLE, and calibration methods to optimize ensemble weighting.