( About ) Iris Vega
AI engineer · Toronto
IMG_0413.RAW · 30PX
Hi, I’m Iris. I make AI boring enough to trust.
I started in physics, got distracted by neural nets in 2016, and have been shipping machine learning to real users ever since — from satellite telemetry to hospital triage.
Today I work independently with two or three teams a year. I usually join when there’s a promising prototype and nobody is sure it will survive real traffic, real data or a real audit. I leave when it does — with the evals, dashboards and docs your team needs to own it.
9
Years in ML
42
Models in production
3.1k
Open-source stars
( 01 ) Experience
Where I’ve been.
Staff AI engineer for hire
Staff engineer, search
ML engineer, speech
ML engineer
( 02 ) Toolbox
What I reach for.
Models
PyTorch
JAX
Hugging Face
vLLM
Retrieval
Vespa
pgvector
BM25 hybrids
Cross-encoders
Infrastructure
Rust
CUDA
Kubernetes
Modal
( 03 ) How I decide
Three rules I don’t break.
01
Measure before you model
If we can’t say what “better” means in a number, we’re not ready to train anything.
02
Small beats clever
A tuned small model you can run and debug beats a giant one you can only hope about.
03
Know when to say no
Sometimes the right answer is a rule, a search box or a person. I’ll tell you when.