Senior AI Training Engineer (AI Systems Engineer) — built SFT/DPO pipelines, sanitized instruction datasets, and ran evaluations
Text
Prompt + Response Writing (SFT)Built automated supervised fine-tuning (SFT) and Direct Preference Optimization (DPO) pipelines to train LLMs on instruction data. Implemented data sanitation steps to remove corrupt tokens, duplicates, and faulty formatting so the training prompts and responses could be ingested correctly. Designed evaluation and consistency checks against gold-standard benchmarks to validate model outputs during training. • Developed SFT and DPO training pipeline components using PyTorch and Hugging Face. • Created preprocessing/sanitation to clean and standardize training instruction sets. • Configured DeepSpeed and mixed-precision routines for efficient parallel model evaluations. • Authored static evaluation scripts and documented engineering guidelines for reproducibility.