Thomas Lefebvre
banner
tlbvr.com
Thomas Lefebvre
@tlbvr.com
Product/ML Eng. I'm building unsplash.com

I write things on tlbvr.com
Montreal 🇨🇦
Pinned
I convinced @unsplash.com to train a vision model for categorizing illustrations.

I've been pushing for this project for a while. I've put together a blog post to share a little bit about our process:

tlbvr.com/blog/categor...
Categorizing illustrations using deep learning: Part 1 | Thomas Lefebvre
How I convinced Unsplash to train a custom neural network to categorize illustrations and improve search.
tlbvr.com
Using k-fold cross-validation has proven useful to find label issues in our dataset. I write a little bit about it here tlbvr.com/blog/k-fold-...
Detect label issues in your vision dataset | Thomas Lefebvre
Leverage cross-validation and identify potential issues in your labeled data for your computer vision model
tlbvr.com
October 3, 2025 at 4:24 PM
I wish python `match` would be an expression. There I said it.
September 23, 2025 at 3:39 PM
Just started watching Stanford's computer vision lectures from this year www.youtube.com/watch?v=2fq9...

see you on the order side.
www.youtube.com
September 21, 2025 at 5:37 PM
Does "reporting junk" do anything on iMessage? My phone number probably hung out in dark places and now i'm flooded with spam. Keep reporting but I don't feel like it does anything.
September 10, 2025 at 3:12 PM
Lowering the loss of your model feels like a game. You initially take bets and get rewarded (maybe).
September 2, 2025 at 5:48 PM
It's that time of year again when I can stick up my feet close to my gaming computer and it keeps them warm.
August 26, 2025 at 4:02 PM
Going from 30s latency to ~400ms in model invocations by dropping the mlflow provided scoring server by a custom one. I don't understand

github.com/mlflow/mlflo...
Latency for served model REST APIs · Issue #7948 · mlflow/mlflow
Willingness to contribute No. I cannot contribute this feature at this time. Proposal Summary For certain requirements, the latency of the model REST APIs have to be very low. We have observed that...
github.com
August 21, 2025 at 5:08 PM
Converting pip groups from pyproject to requirements.txt is sometimes necessary to interop with tooling that don't have pip>=25 or some automation picking up on requirements.txt. I've been using a script provided by the python docs that does just that.

packaging.python.org/en/latest/sp...
Dependency Groups - Python Packaging User Guide
packaging.python.org
August 12, 2025 at 5:06 PM
AI-powered IDEs coming out left and right...as often as...JS frameworks?
August 6, 2025 at 2:02 PM
Coming from a serious Typescript and other typed languages background, I was a bit worried working in python but I'm pleasantly surprised about the ecosystem revolving around types. Finding separate type stubs reminds me of the early TS days where most libraries required custom .d.ts files.
August 4, 2025 at 1:45 PM
I'm looking into ways of spotting labels issues in my training data. Cleanlab github.com/cleanlab/cle... has a feature where you feed predictions and it'll tell you which image may have a bad label based on the confidence of the predictions.
GitHub - cleanlab/cleanlab: Cleanlab's open-source library is the standard data-centric AI package for data quality and machine learning with messy, real-world data and labels.
Cleanlab's open-source library is the standard data-centric AI package for data quality and machine learning with messy, real-world data and labels. - cleanlab/cleanlab
github.com
July 28, 2025 at 11:34 PM
Friday night. Will this job pass before I pass out?
July 25, 2025 at 9:56 PM
I live in the MLFlow dashboard these days. Let's see if `likeable-horse-161` performs better.
July 23, 2025 at 3:31 PM
I'm preparing a presentation for my team @unsplash.com to walk them through our recent progress in categorizing illustrations

If you missed it, I wrote a blog post about it bsky.app/profile/tlbv...
July 9, 2025 at 9:15 PM
I convinced @unsplash.com to train a vision model for categorizing illustrations.

I've been pushing for this project for a while. I've put together a blog post to share a little bit about our process:

tlbvr.com/blog/categor...
Categorizing illustrations using deep learning: Part 1 | Thomas Lefebvre
How I convinced Unsplash to train a custom neural network to categorize illustrations and improve search.
tlbvr.com
July 2, 2025 at 12:50 PM
I'm writing a new blog post. Lil' sneak peak, here's the title: "Categorizing illustrations using a convolutional neural network". Might take a bit to get it out. Bonus, it features @kirillz.com
June 14, 2025 at 10:00 PM
When your day is going well, very much know what needs to be done.
June 5, 2025 at 2:59 PM
I dont know how folks can even read news on their phone nowadays
April 30, 2025 at 4:23 PM
I'm going to go back to Motherland belgium for a couple of days. I'm looking at surprising my dad in a nice restaurant. I haven't been to belgium in a long time so I don't know what's good. If anyone knows any recos in the Mons region, send them my way!
February 24, 2025 at 4:22 PM
Saturday morning maintenance?
February 22, 2025 at 1:37 PM
I just live by this command nowadays.

`darwin-rebuild switch`
February 14, 2025 at 1:49 PM
First time in my life I’ve been looking at aircrafts before booking a flight. Too many incidents with Boeing lately.
January 2, 2025 at 5:30 PM
Loaded up on coffee and pushed my first post of this year tlbvr.com/blog/benchma... — If you're into semantic search, this one might be of interest. Happy new year.
Benchmarking Embedding Models for Semantic Search | Thomas Lefebvre
Explore how to benchmark embedding models to optimize restaurant discovery using semantic search.
tlbvr.com
January 2, 2025 at 2:47 PM
As promised, here's the jupyter nb github.com/Magellol/ben... — I'm working on a blog post as we speak.
December 30, 2024 at 10:58 PM