Michael Jaynes

I am a Senior AI Engineer focused on model evaluation, embedding pipelines, and stabilizing distributed training workflows for large language models.

At Meta I debugged LLaMA-based systems, inspected token-level divergences, resolved CUDA/PyTorch DDP instability, and tracked embedding drift using FAISS and Python tooling.

Previously at Google I developed large-scale preprocessing and training pipelines with Python, Apache Beam, and JAX/Flax on TPU pods, built gRPC model-serving components in Go, and implemented caching and monitoring layers to stabilize inference latency.

I build developer-facing tools in TypeScript and React to surface token-by-token comparisons, reasoning traces, and output-drift visualizations, and I prioritize reproducible evaluation suites, rollout validation, and robust infra debugging.