Krishna Mahajan

Experience

1. Machine Learning Engineer - Ml Platform

   Amazon Fulfillment Technologies & Robotics

   Mar 2025 — Present · Bellevue, WA, US

   Trained 7B Qwen 3 VLM using Distributed Data Parallel (DDP / ZeRO-0), optimizing for throughput, simplicity, and rapid iteration- Scaled to ~32B parameter models (Qwen-32B) using ZeRO-3 in LLaMA-Factory- Led training of ~300B parameter models(GLM 4.5/GLM 4.6) using MS-Swift/Megatron-Bridge, leveraging multi-dimensional model parallelism- Developed Expertise in selecting parallelism strategies based on model size, memory pressure, and communication overhead- Optimized training systems to support ultra-long context lengths (up to ~30K tokens)- Collaborate closely with ML researchers to translate research requirements into scalable training configurations, enabling experimentation at previously infeasible model sizes- Architected and owned three decoupled, production-grade ML pipelines (Tokenization, Training, Continuous Evaluation), designed for long-running, large-scale workloads with strong modularity and low coupling- Orchestrated multi-node, multi-GPU distributed training on AWS P5 (H100) clusters, implementing custom master-node discovery, rank coordination, and synchronization barriers- Designed and built a continuous evaluation engine that monitors S3 checkpoints and triggers decoupled inference jobs, enabling real-time validation without impacting training throughput- Engineered streaming-first data ingestion using MosaicML StreamingDataset (MDS), enabling terabyte-scale training directly from S3 - Built and optimized production grade Docker images for H100 instances for all workloads- Integrated end-to-end experiment tracking with Weights & Biases, designing grouped-run workflows to correlate metrics across decoupled pipelines.

Education

• Indiana University Bloomington

  Master’s Degree, Data Science

• Dhirubhai Ambani University

  Bachelor of Technology (BTech), Computer Science

Skills

• Java
• Data Analysis
• Data Structures
• Css
• Statistical Data Analysis
• Data Mining
• Microsoft Excel
• Hadoop
• R
• Javascript
• Html
• Etl Tools
• Hive
• English
• Python
• Algorithm Design
• Data Analytics
• Statistics
• Machine Learning
• Sql
• Linux
• Microsoft Office