live from the lab

Big models, small footprint.

Wes runs AI at home: large language models, image pipelines and agent experiments on a desk-sized supercomputer. If it can run locally, it runs locally.

DGX Spark

online
gpu
38%
throughput
41.8tok/s
temp
54°C
weights 38.4 GBkv cache 20.5 GBfree of 128 GB

The box

spec
chip
GB10 Grace Blackwell
memory
128 GB unified
ai perf
up to 1 petaFLOP FP4
form
fits next to a coffee mug
client
MacBook Pro, M5 Max

Local LLMs

ollama
served on devicetokens 0rate 0.0 tok/s

Image pipeline

comfyui
step 0 / 30sampler eulercfg 6.5seed 0

Trajectory Labs

studio

Wes's product studio. A portfolio of small, focused apps and developer tools that each pay their own way.

DishDecoded

in development

An iOS nutrition tracker that uses AI to make logging meals fast. Flutter on the front, Supabase behind.