Aashrith Sai

About Aashrith Sai

Experienced AI/ML Research Engineer with a strong background in leveraging large language models (LLMs) to enhance system capabilities and streamline data processing. Proven expertise in developing continuous learning pipelines, deploying AI systems efficiently, and improving data quality and system performance. Adept at designing and implementing end-to-end diagnostics pipelines and contributing to production-level systems. Passionate about leveraging machine learning to solve complex problems and drive innovation.

Experience

1. Machine Learning Engineer Ii

   Amazon

   May 2025 — Present

   Improved long-context LLM inference throughput by 30–45%(32 K–128 K tokens) by optimizing batch scheduling, resolving CUDA bottlenecks, and tuning tensor-parallel attention in vLLM and TensorRT, enablingstable large-sequence performance at enterprise scale.2. Enhanced evaluation infrastructure and pipeline reliability (~ 35%) by restructuring Airflow DAGs for nightly and online evaluations, adding fallback and retry logic, and hardening failure-recovery paths supporting reliable benchmarking across internal and external models (including Nova/Bedrock-backed models).3. Stabilized multi-node inference deployments for vLLM/TRT and Nova-based workloads by addressing KV cache growth, memory fragmentation, throttling, and timeout handling — reducing reruns and mid-run interruptions by ~ 25% under production-scale load.4. Expanded model evaluation and deployment coverage by 4×, integrating advanced reasoning, long-context recall, graph and agentic benchmarks into a unified evaluation framework that supported publicly released models like Amazon Nova — accelerating model comparison cycles by ~ 40% and helping deliver validation results at the release-ready level.

Education

• Chaitanya Bharathi Institute Of Technology

  Bachelor of Engineering - BE

• Texas A&M University

  Master of Science - MS