LinkedIn Cringe Analyser

How I built it

Backend (Machine Learning):

I scraped thousands of posts from the LinkedInLunatics subreddit, cleaned and processed the data, and built a custom LightGBM model paired with BERT tokenizers to extract features like sentiment, buzzwords, and cringe patterns. The API server is built using FastAPI, which handles model inference and delivers results at lightning speed.

Frontend:

The user interface is built with React and enhanced with Three.js for a visually engaging and interactive experience. Users can simply type their post, submit it, and instantly get a score with feedback.

Deployment:

The entire project—backend, frontend, and ML model—is deployed on Render, ensuring it’s accessible anytime.

Challenges I ran into

• Cleaning messy, unstructured data from LinkedInLunatics required a lot of custom preprocessing using OCR.
• Balancing the lightweight speed of LightGBM with the powerful feature extraction of BERT took some fine-tuning, I tried using GPT but too much of latency.
• Designing a fun and engaging interface with Three.js while keeping it simple and user-friendly was a creative challenge.

Accomplishments that I'm proud of

• Building a functional and kinda accurate ML-powered cringe analyzer entirely on my own.
• Creating a seamless user experience with an interactive UI and fast backend responses.
• Deploying the project successfully on Render and making it available for everyone to try.

What I learned

• Oh boy oh boy, scraping and cleaning data can be just as challenging as building the model itself.
• Combining traditional ML tools like LightGBM with modern NLP techniques like BERT can yield efficient and accurate results.
• Balancing backend performance with a fun and interactive frontend is a rewarding but time-consuming task.

What's next for LinkedIn Cringe Analyzer

• Adding GPT-powered suggestions to help users improve (or embrace) their cringe factor.
• Expanding the dataset to include cringe-worthy content from platforms like Twitter or Instagram.
• Developing a real-time cringe analysis browser plugin for LinkedIn posts.
• Enhancing the front-end experience with more animations and features, like a "most cringe leaderboard."

Building this project solo was an incredible experience, and I hope you enjoy using it as much as I enjoyed making it!

Built With

• bert
• fastapi
• lightgbm
• numpy
• pandas
• praw-(reddit-api)
• pydantic
• python
• pytorch
• react
• scikit-learn
• spacy
• tailwind
• tesseract-ocr
• textblob
• three.js