This extension helps you keep your LinkedIn feed professional.
I've seen one too many memes with Yusuf Dikec for my own good in the past week. From attention grabbing to sales and product marketing, it's been adding a lot of noise.
So I've created LinkedOut.
The project has two parts:
- A Python script that exports a pre-trained neural network to the ONNX format
- A Chrome extension that filters LinkedIn feed posts
The Python script requires Python 3.10 and Poetry to be installed. Oh, and the Torch versions are for MacOS only. You're better off just using models/model.onnx
, which is an exported version of EfficientNetB0.
The Chrome extension loads the model in ONNX format via onnxruntime-web
and computes embeddings for:
- User-uploaded images, in the extension popup.
- Posts on the LinkedIn feed, while you are scrolling.
If the similarity between the image of a LinkedIn post and any of the user-uploaded images is greater than a threshold, it removes the post from the feed.
For the extension, you can build it locally by doing
npm run build
To use it in Chrome:
- Navigate to
chrome://extensions
- Enabling
Developer Mode
- Click
Load unpacked
- Point to the
dist/
folder in the extension repo, which was generated by thenpm run build
command.
If you want to use the Python script:
> poetry env use python3.10
> poetry install
> poetry shell
> (image-embedding-py3.10) python model.py
And then your model should be exported under models/model.onnx
.
Q: Does it work?
Not really. The similarity scores are weird sometimes, I've had both false positives and false negatives.
Q: Does it impact my Chrome experience?
You will probably have small delay/lag when loading your LinkedIn feed. This also depends on your resoruces.
Q: Is this extension production-ready?
No, I've just scaffolded it in a few hours with ChatGPT. I can conut on the fingers from my right hand how many times I have written JS/TS before. Apologies if the code makes your eyes bleed.
Q: Can I fork this and do whatever I want?
Yes, as long as you're not making money from it.
Q: Why no Docker for the Python thing?
🤷