Rajia Rani

About Rajia Rani

Hello, I’m I come from a, transitioning from.For +, I taught Chemistry and mentored students in Math and Science. That’s where I learned how to, think deeply, and solve problems step by step.In 2023, I transitioned into - and never looked back.Today, I work as a @USD, building and experimenting with:• Large Language Models (LLMs) and AI Agents• PyTorch and GPU-based trainingI enjoy turning complex ideas into and continuously improving how they perform.Alongside this, I’m pursuing a ’ to strengthen my research and technical foundation.

Experience

1. Research Assistant

   Usd Artificial Intelligence ᗩ Research

   Mar 2026 — Present · US

   Implemented an LLM pre-training pipeline using Megatron-DeepSpeed with 5D parallelism (DP+TP+PP+CP/SP+EP) and ZeRO(stage 2/3) to scale training across 8 GPUs; improved tokens/sec by 28% while maintaining 85% scaling efficiency.2. Enabled long-context training using sequence/context parallelism and activation recomputation; extended sequence length from 2K to 8K tokens with 35% lower peak memory usage and 18% throughput improvement.3. Developed reproducible experiment pipelines (configs, scripts, benchmarking workflows); reduced experiment setup and reproducibility time from 2 days to 4–6 hours across GPU environments.

Education

• Govt Girls Sr.Sec School Nehru Garden Jalandhar

  12, Science (Non-Medical)

• University of South Dakota

  Master of Science - Artificial Intelligence

• Lovely Professional University

  Master of Science - MS, Chemistry

• PCM SD College for Women Jalandhar

  Bachelor's of Science , (Non-Medical)