Founding ML Engineer

github

Dev Khant  • Peerlist

• Here for an given image with a car number plate is detected and the number is extracted.
• For detecting plate YOLOv4 is used and used AWS Textract to extract number from cropped image.
• Flask is used to build the web app and Docker for containerizing the app.
• Dockerized flask app can be deployed on Heroku, AWS, Google Cloud, or Azure.

Technology used :  Python, Tensorflow, OpenCV, YOLO, HTML, CSS, Flask, Docker, AWS

Car Number Plate Detection & Extraction

• Here I built a classifier for Cars and Bikes. Data was collected by myself from real world. 
• I used Transfer Learning and the model was EfficientNet. 
• The final accuracy was 99.2%.

• Technology & Libraries : Deep Learning, Transfer Learning, Tensorflow, EfficientNet, Sklearn

Cars Vs Bikes Classifier

• Developed a Chrome extension to get a detailed summary of a GitHub Repository using LLM.
• It uses Langchain to interact with gpt-3.5-turbo from OpenAI. Flask is used as a backend server to handle the requests.
• It traverses through each code file in the repository and stores the summary of it. At the end, model summarizes all the
stored summaries into one final response.

• Learned how to use LLM using Langchain. And also how to build a web extension.

Github Code Analyzer

• It is about explaining in text about the image. I used EfficientNet for image processing and GRU for text generation. 
• Attention mechanism is also used to get focus on certain part of image. Dataset used here was Flickr(30K).
• BLEU score obtained by the model was 13%.

• Technology & Libraries : Python, Computer Vision, NLP, Tensorflow, NLTK, Pandas, Numpy

Image Captioning

• This project is about translating English sentences to Hindi sentences using Transformers from scratch.
• It is implemented using Tensorflow and IIT Bombay Translation Dataset was used to train the model. 
• The BLEU score obtained by the model is 25%.

• Technology & Libraries : Python, Deep Learning, NLP, Transformers, Tensorflow, NLTK

Machine Translation using Transformers

• This is about checking whether a review is positive or negative. 
• I used Support Vector Machine(SVM) to build the classifier using Sklearn. 
• The accuracy for the model was 88%.

• Technology & Libraries : Python, Machine Learning, SVM, NLTK, Flask, HTML, CSS, JS, Heroku