Writing & appearances

Greg and I wrote this around four patterns we teach: prompt engineering, RAG, fine-tuning, and agentic reasoning. A guide to building and shipping LLM applications.

550B-parameter MoE hybrid Mamba-Transformer with 1M-token context, built for high-throughput agentic reasoning.

120B (12B active) LatentMoE hybrid Mamba-Attention model — first in the family pre-trained in NVFP4, with MTP layers for native speculative decoding.

Technical report for the Nemotron 3 family — MoE hybrid Mamba-Transformer models (Nano, Super, Ultra) with NVFP4 training, LatentMoE, and 1M-token context.

Upcoming — co-located summits for engineers building industry-ready agentic and generative AI.

Breaking down the 550B hybrid Mamba-Transformer MoE and what it means for long-running agents.

I host conversations and Q&As with the people building Nemotron on the NVIDIA AI channel.

The bootcamp I co-created and taught: prompt engineering, RAG, fine-tuning, agents, and evals. The curriculum became our book (Wiley, 2026).

Videos on LLMs, fine-tuning (LoRA and friends), and ML techniques.

A back catalog of live sessions I co-hosted on LLMs, agents, evals, and AI engineering tools.