This portfolio consists of a compilation of projects that I created for various purposes, demonstrating my skills and experiences in different areas.
- The Impact of AI on the Practice of Law: Opportunities, Challenges, and Legal Implications
- How to Develop an AI Career as a Beginner
- Learning How to Learn: Techniques to Master Any Difficult Subject Effectively
- How to Continuously Improve AI Models with Changing Data
- The Blueprint for Building Successful AI Products
- How to split image dataset into train, validation and test set?
- How to Detect and Remove Corrupted Image Files in an Image Dataset Using Python ?
- How to use Kaggle datasets in Google Colab?
- Nonlinearity and Neural Networks
- Impact of AI on Social Engineering
- Fastai Car Type Classification Model Using a Custom Dataset
- How to Deploy Fast.ai Models? (Voilà , Binder and Heroku)
- Computer Vision Archetypal Problems
Tea Recommendation Chatbot
Finding the perfect tea can be overwhelming with so many options available. Consumers often struggle to find teas that match their taste preferences, health goals, and emotional needs, making the process frustrating and time-consuming. This chatbot simplifies tea selection by recommending the perfect tea for you from over 100 types. It provides personalized suggestions based on your individual preferences, helping you find the ideal tea with ease.
-
Technologies Utilized: RAG, LLM, Lama-3.1-70B, GPT-4, prompt engineering
Targeted Social Media Marketing Campaign
Use data and messaging tools to reach a specific audience.
- Start with a dataset containing contact details (e.g., phone numbers, emails, addresses, age).
- Identify a target audience based on geographic location and demographics.
- Clean and analyze the data to focus on the relevant audience.
- Extract contact details (phone numbers, emails) for the selected group.
- Use a bulk SMS API to send personalized messages and emails to the targeted audience.
- Tools: Python, Pandas, Data Cleaning & Analysis, SMS API.
Translating PDF e-book to English using OCR and LLMs
This project aims to translate a Sinhala language PDF e-book into English using Optical Character Recognition (OCR) and Large Language Models (LLMs).
- Convert PDF to Images: Convert each page of the Sinhala PDF e-book into an image.
- Extract Text with OCR: Use an OCR model (e.g., Tesseract) to extract text from each image.
- Translate with LLM: Use GPT-4 or similar LLM for translating the extracted Sinhala text into English.
- Workflow: PDF → Images → OCR → LLM (Translation).
- Tools: OCR (Tesseract), LLM (GPT-4), Python, Prompt Engineering.
Fine-Tuning Stable Diffusion for Personalized Image Generation
Engineered prompts and utilized stable diffusion techniques to fine-tune the model for generating personalized images.
- Applied prompt engineering methodologies to tailor the stable diffusion model for personalized image generation.
- Leveraged stable diffusion techniques to ensure stable and high-quality image generation results.
- Implemented the solution using Python programming language.
- Technologies Utilized: Prompt Engineering, Stable Diffusion, Python
Chatbot Development with ChatGPT API
Developed a personalized chatbot with persona using the ChatGPT API.
-
Implemented prompt engineering techniques to customize the chatbot's responses based on user inputs and predefined personas.
-
Utilized the ChatGPT API for natural language processing and response generation.
-
Developed the solution using Python programming language.
-
Technologies Utilized: ChatGPT, Prompt Engineering, Python
-
Joker Chatbot - An AI-powered conversational agent inspired by the character Joker from The Dark Knight.
-
Morpheus Chatbot - An AI-powered conversational agent inspired by the character Morpheus from The Matrix.
AI-Assisted Data Labeling
Implemented AI models to assist human labelers, enhancing labeling efficiency and accuracy.
- Trained AI models to provide initial labels, minimizing the need for human labelers to start from scratch.
- Human labelers only need to correct mistakes made by the AI model, streamlining the labeling process.
- Conducted training sessions for labelers to ensure they understand and effectively utilize the AI-assisted labeling system.
- Established labeling standards, setup labeling tools, and performed quality checks to maintain labeling accuracy.
- Utilized Keras, TensorFlow, OpenCV, Python, and various techniques such as image recognition, object detection, OCR, and Label Studio for efficient AI-assisted data labeling.
- Technologies Utilized: Keras, TensorFlow, OpenCV, Python, Image Recognition, Object Detection, OCR, Label Studio
Medical Document Patient Information Retrieval using LLM
Implemented a system to extract patient information from medical documents using LLM.
- Utilized BaiduOCR for initial OCR results of medical document images.
- Leveraged LLAM2, llamaindex, and Retrieval Augmented Generation (RAG) techniques for accurate extraction of patient information from medical document images.
- Technologies Utilized: RAG, LLM, LLAMA2, llamaindex
Number Plate Detection and Recognition
Designed and implemented a Chinese number plate detection and recognition system to extract number plate information from car images.
- Conducted in-house data collection, established labeling standards, and trained data labelers to ensure accurate labeling and maintain quality.
- Created a comprehensive dataset of number plates for model training.
- Developed a number plate detection model using labeled data.
- Implemented a two-step process: first, utilized the detection model to locate number plate positions, then employed the Baidu OCR model for number recognition.
- Developed a demonstration to showcase the system's functionality.
- Successfully deployed the number plate detection and recognition pipeline to extract number plate information from images.
- Technologies Utilized: CNN, YOLO, OCR, Keras, TensorFlow, OpenCV, Python, Image Recognition, Object Detection
Damage Car Part Detection/Recognition
Developed a series of computer vision AI models to automatically detect and extract information about damaged car parts from car accident images.
- Conducted in-house data collection and established labeling standards. Trained data labelers and conducted quality checks to ensure accurate labeling.
- Created datasets for damage detection and recognition.
- Developed various AI models to automatically extract information about damaged car parts, including:
- Detection and recognition of damaged and undamaged parts.
- Classification of damaged severity (e.g., destroyed, scratched, deformed, bent).
- Recognition of repair or replacement needs for car parts.
- Detection of damaged areas within car images.
- Created a demo to showcase the functionality of the developed models.
- Successfully deployed these models in operational settings for information extraction from images.
- Technologies Utilized: CNN, YOLO, Keras, TensorFlow, OpenCV, Python, Image Recognition, Object Detection, LabelStudio
Car Part Detection
Developed a visible car part detection model to identify visible car parts in car images.
- Conducted in-house data collection, established labeling standards, and trained data labelers for accurate labeling.
- Utilized YOLO, Keras, TensorFlow, and OpenCV for model development.
- Successfully deployed the model to detect visible car part information from images in operational settings.
- Technologies Utilized: YOLO, Keras, TensorFlow, OpenCV, Python, Object Detection
Car Part Segmentation
Developed car part segmentation models to segment visible car parts in images.
-
Conducted in-house data collection, established labeling standards, and trained data labelers to ensure accurate labeling.
-
Built car part segmentation models using the UNet architecture and image segmentation techniques.
-
Utilized OpenCV, Python, TensorFlow, and Keras for model development.
-
Successfully deployed the models to segment visible car parts in images.
-
Technologies Utilized: UNet, Image Segmentation, OpenCV, Python, TensorFlow, Keras
Image Categorizers
Developed various classifier models to automatically categorize and process images.
-
Conducted in-house data collection, established labeling standards, and trained data labelers to ensure accurate labeling and quality control.
-
Built classifiers to categorize images into different types such as car images, car components, documents, driver images, and car interiors.
-
Developed specific classifiers for
-
Car brand type classification
-
Car model type classification
-
Car viewpoint type classification (e.g., front, back, right side, left side)
-
Car type classification (e.g., sedan, truck, jeep)
-
Document type classification (e.g., ID card, driver's license, vehicle license, VIN, bank card)
-
Car color classification
-
-
Technologies Utilized: CNN, Keras, TensorFlow, OpenCV, Python, Image Recognition
Image Grouping by Car
Developed a system to separate and group jumbled images belonging to multiple cars into distinct sets for each car.
- Utilized image feature clustering, number plate recognition, and car color recognition to group images.
- Employed CNN for image feature extraction and KNN for image feature clustering.
- Integrated number plate recognition to accurately identify and group images of the same car.
- Implemented car color recognition to enhance grouping accuracy.
- Successfully separated and grouped images of each car using the developed system.
- Technologies Utilized: YOLO, CNN, KNN, Clustering, Image Recognition, Object Detection
Document Classification and OCR
Developed a system for document classification and OCR.
- Implemented document type classification to differentiate between ID cards, driver's licenses, vehicle licenses, VINs, and bank cards.
- Built a detection model to identify and isolate various document types.
- Separated different types of documents and utilized Baidu OCR to extract information from them.
- Technologies Utilized: CNN, Keras, TensorFlow, OpenCV, Python, Image Classification, YOLO, Object Detection, BaiduOCR
Document Orientation Detection and Correction
Developed a system capable of detecting the orientation of documents and rectifying any incorrectly rotated ones. The model can detect four orientation angles: 0°, 90°, 180°, and 270°.
- Employed CNN models implemented with Keras and TensorFlow for precise document orientation detection.
- Utilized OpenCV for efficient image processing and correction of improperly rotated documents.
- Successfully integrated the system to accurately detect and rectify document orientations.
- Technologies Utilized: CNN, Keras, TensorFlow, OpenCV, Python, Image Recognition
Car Wheel Alignment Screen Image OCR
Car wheel alignment machines are essential tools in auto repair shops, providing precise measurements of a vehicle's wheel alignment. The screen display of these machines typically shows various key parameters that technicians use to adjust the alignment of the wheels. The goal of this project is to extract parameters from wheel alignment machine screen images. We have developed a comprehensive system for OCR (Optical Character Recognition) of car wheel alignment screen images.
- Conducted wheel alignment screen type classification to categorize different types of screens accurately.
- Implemented OCR techniques to extract wheel alignment information from various types of wheel alignment screen images obtained from repair shops.
- Applied for a Chinese patent for the developed system, with the patent result currently pending.
- Technologies Utilized: CNN, Keras, TensorFlow, OpenCV, Python, Image Classification, Object Detection, OCR
Medical Document Classification and OCR
Developed a comprehensive system for medical document classification and OCR.
- Implemented a medical image type classifier to distinguish between different types of medical documents such as patient medical records and diagnosis proofs. Utilized keyword search for further classification.
Waste Management Using Blockchain
Developed a project proposal outlining the exploration of blockchain integration in waste management.
- Technologies Utilized: Blockchain
