Portfolio

LLM for Document Based Question Answering - Github Repository

Skills: Large Language Models | RAG Implementation | Prompt Engineering | Chat Automation

Project Overview:

• Developed Large Language Model (LLM) using RAG implementation to interpret PDFs, CSV, and JSON files, with 98% accuracy.
• Used LangChain for efficient storage and retrieval, ensuring accurate responses by implementing Prompt Engineering.
• Enhanced chat automation processes by testing model performance on diverse and complex queries.

Smart Campus Human Detection

Skills: Deep Learning

Project Overview:

• Annotated a dataset used for a smart campus project involving human detection with a 3D LiDAR camera.
• Developed and tested a Neural Network AI model using 2D and 3D CNNs to detect human presence, achieving 98% accuracy.
• Contributed to the development of a smart campus solution aimed at monitoring and managing crowd density in campus areas.

Skin Disease Detector App - Github Repository             

Skills: Deep Learning | Computer Vision | Real-Time Classification | Model Optimization | Data Augmentation

Project Overview:

• Led the development of skin disease detector app during a 24-hour hackathon, using TensorFlow Lite for real-time classification.
• Preprocessed and augmented Harvard skin disease images using OpenCV and did model tuning to achieve 82% accuracy.
• Developed and integrated the model into an Android app for disease classification via smartphone camera and to find nearest dermatologist.

Brain Tumor Recognizer App - Github Repository

Skills: Model Optimization | Azure Custom Vision | Data Augmentation

Project Overview:

• Developed AI model for real-time image classification of 2-D brain MRI scans using TensorFlow Lite and Azure’s Custom Vision Service.
• Preprocessed and augmented Harvard dataset skin disease images with noise addition and model tuning for optimized performance.
• Developed and exported a TensorFlow Lite model, integrated it into an Android app for mobile healthcare via smartphone camera.

Satellite Image Segmentation with SLIC and Canny

Skills: SLIC | Canny Edge Detection | HOG | GLCM | K-Means Clustering | Remote Sensing

Project Overview:

• Combined Simple Linear Iterative Clustering (SLIC) and Canny Edge Detection for precise segmentation of satellite images.
• Utilized Histogram of Oriented Gradients (HOG) and Gray-Level Co-occurrence Matrix (GLCM) to refine superpixels and enhance boundary precision.
• Applied K-means clustering and label mapping to achieve accurate classification in remote sensing imagery.

Gesture-Based Media Control System - Github Repository

Skills: Deep Learning | Data Augmentation

Project Overview:

• Curated a dataset of hand gestures for play/pause, volume up, and volume down functions.
• Applied preprocessing and augmentation techniques to enhance gesture recognition accuracy to 95%.
• Developed a gesture recognition model and integrated it with media controls for hands-free operation.

Material Search Engine

Skills: Machine Learning | Deep Learning

Project Overview:

• Led my team to develop a classification model for material detection using d-values from X-ray diffraction data.
• Trained and evaluated various classification models to identify material types based on 4 distinct d-values with ∼100% accuracy.

EV Battery RUL Predictor - Github Repository

Skills: Machine Learning | Django | NASA Prognostic Dataset | Data Visualization | Predictive Analytics

Project Overview:

• Developed a machine learning solution to predict the Remaining Useful Life (RUL) of EV batteries.
• Integrated prediction models into a Django-based website for user-friendly inputs and automated predictions.
• Visualized battery degradation using the NASA Prognostic dataset to track performance over time.
• Presented real-time predictions for battery performance and life expectancy, enhancing user decision-making on battery usage.

Bus Number Recognition System - Github Repository

Skills: Computer Vision | OCR | Google Cloud Compute | Arduino Integration | Real-Time Data

Project Overview:

• Developed a bus number detection app using OpenCV and EasyOCR to identify and announce bus numbers for visually impaired users.
• Set up a Google Cloud Compute server and integrated it with an Arduino client to capture and process images.
• Integrated Arduino with Singapore government systems to fetch real-time bus arrival data and provide announcements.

Super Resolution on Jetson Nano

Skills: Super Resolution | Machine Learning | PyTorch | C++ | ARM-based Systems

Project Overview:

• Implemented a super-resolution model on an ARM-based Jetson Nano device for real-time photobooth applications.
• Quantized the model to optimize performance on the low-memory Jetson Nano, ensuring efficient computation.
• Integrated C++ with PyTorch for faster inference and improved processing speed.

Process Scheduling Visualization Application - Github Repository

Skills: Java | Data Visualization | Process Scheduling | GUI Development

Project Overview:

• Developed a Java app to compute imoprtant metrics for process scheduling algorithms.
• Implemented features for input and visualization of process data, including options to add, edit, and execute scheduling algorithms.
• Generated visual graphs to represent calculated metrics, enhancing user understanding of scheduling performance.

Student Drowsiness Detection System - Github Repository

Skills: Deep Learning | Computer Vision | Data Extraction | Real-Time Prediction

Project Overview:

• Extracted 57,488 images from ultraLDD dataset, applied masking using dlib, faceutils along-with OpenCV to focus on eyes and mouth.
• Optimized data handling with a custom TensorFlow data generator for memory-efficient CNN training, achieving 93% accuracy.
• Implemented real-time prediction from live camera feeds or saved videos, displaying drowsiness detection results with overlay on the screen.

Pore Pressure Gradient Prediction

Skills: Machine Learning | Deep Learning | Data Visualization

Project Overview:

• Utilized machine learning and deep learning to forecast Pore Pressure Gradient (PPG) for optimal oil well drilling with an RMSE of 0.15.
• Validated model predictions, ensuring accuracy and practical applicability minimizing the need for dynamite blasts.
• Fine-tuned various regression and deep learning models, successfully selecting XGBoost for superior prediction accuracy and efficiency.

Node Classification in Graph Networks Using GNNs

Skills: Deep Learning | Graph Neural Networks (GNN)

Project Overview:

• Developed and optimized Graph Neural Network models (GCN, GAT, GraphSAGE) achieving accuracy of 90.16% on the Cora dataset.
• Implemented and tested advanced techniques like attention mechanisms and scalable embeddings to improve classification accuracy.
• Collaborated in a team to analyze results and present findings, highlighting GNN applications in real-world scenarios.

Smart Parking Database Management System - Github Repository

Skills: Database Design | Data Testing | SQL | Data Validation

Project Overview:

• Led development of comprehensive database management system for smart parking lot application.
• Generated 24,000 entries of dummy data to thoroughly test and validate the functionality of the database management system.
• Created comprehensive documentation with SQL queries, query output snapshots, and CSV exports for database testing and analysis.

Twitter Sentiment Analysis - Github Repository

Skills: Natural Language Processing (NLP) | Machine Learning | Data Visualization | Data Pipelines

Project Overview:

• Developed Twitter Sentiment Analysis framework, utilizing TweePy and incorporating Natural Language Processing (NLP) techniques.
• Engineered data pipelines to analyze over 30,000 tweets, revealing sentiment trends through word clouds, pie chart and scatterplots.
• Applied K-Nearest Neighbors (KNN) classifier, enhancing the analysis of sentiment distributions across different political groups.

Graph Operations and Search - Github Repository

Skills: Java | Graph Operations | CLI Development

Project Overview:

• Developed Java application for graph operations, with features like parsing DOT files, node and edge management, and visualization.
• Developed and integrated search algorithms including Breadth-First Search (BFS), Depth-First Search (DFS).
• Provided command-line interface for interacting with graph structures, performing searches, and exporting results in .dot and image formats.

Temporal Hyperlink Prediction

Skills: Machine Learning | Deep Learning | Temporal Prediction | Bibliometric Analysis

Project Overview:

• Conducted research on temporal hyperlink prediction in hypergraph using a bibliometric dataset of 50,000 neuroblastoma publications.
• Enhanced existing mathematical, machine learning, and deep learning techniques of graph link prediction for hypergraph link prediction.
• Developed predictive models with 77% accuracy to forecast emerging research branches and predict recipes from available ingredients.