Martin Sawtell

LLM / Knowledge Graphs / Consulting

Every Engagement Gets a Graph

Four client knowledge graphs in five months. Local extraction on my own GPUs, agentic research and writing chains on top, and the eval discipline that's slowly turning it into a repeatable consulting tool.

Continue reading →
Gaussian splat capture of juicebox, an open-frame GPU rig behind black mesh on a plywood bench in the basement, keyboard and mouse in front, cabling everywhere Automation / LLM / homelab / harness

The Harness (Agents Running the Homelab)

Claude Code as lead engineer over a crew of specialised agents, running on a GPU rig in the basement. Hub-and-spoke delegation, cost rules enforced by hooks, and an agent whose only job is answering What Would Martin Do.

Four panelists in armchairs on the Dell Technologies World 2025 stage under blue arch lighting, Martin far right in a flat cap Space / Digital twins

To the Moon, to Stay (DTW 2025)

A panel at Dell Tech World with the MIT Media Lab crew whose payloads landed near the lunar south pole in March 2025, and how a megabyte from the Moon becomes a gigabyte of terrain.

On the Dell Technologies World 2024 demo floor with Mark Hammond, the screens behind showing his live camera feed restyled into a cat in space Generative AI / Virtual production

15 Cats Per Second (Realtime Diffusion at Dell)

Stable Diffusion running on a live camera feed at 15 frames a second, or as we pitched it on stage, 15 cats per second. TouchDesigner, three kinds of camera, and a LoRA trained on 60,000 selfies.

On the AWE Asia main stage, mid-talk, with a photo of a 2003 live-visuals mixing desk on the screen behind VR / Live performance

Performing in the Metaverse (AWE Asia 2023)

A talk on VRChat's grassroots gig scene- full body tracking, decompiled Half-Life stages, and why 75,000 people a day show up to dance as a scientist with a bald head.

A NeRF-rendered novel view of a person holding a tennis racquet, reconstructed from captured footage Neural rendering / R&D

NeRF'd

Training a neural radiance field from a phone-quality SFM pointcloud, then figuring out what to do with the volume it gives you back.

See all 12 case studies →

The Journal

September 2, 2020 · archive

Eye Tracking Personalities

A video on what we found looking into how different personalities and demographics interact with a visual field test.

August 9, 2019 · archive

2019 Reel Refresh

A quick note setting the 2019 VFX reel free after sitting on it too long.

March 31, 2017 · archive

RfM Installation Docs

Notes and forum correspondence on the fight to get Renderman For Maya and Alfred talking to Maya properly.

March 8, 2016 · archive

Medical Data in Minecraft

Turning volumetric medical data into Minecraft voxels, because the environments are just a bunch of voxels already.

September 12, 2014 · archive

Dynamic camera mapping

A system for dynamically mapping video textures onto geometry in Houdini as it starts moving, tracking changing topology through a cloth solver.

April 22, 2010 · archive

Plastic deformation system

A Houdini 10 workflow for smashing a car into an object using the cloth system's plastic deformation.

March 3, 2010 · archive

PRman and Houdini 10

Getting PRman 14.3 talking to Houdini 10 on OSX 10.5, a combination nobody else had written up at the time.

All 11 journal entries →