Developed a multimodal system combining Vision Transformer (ViT) and GPT-2 for automated radiology report generation. Leveraged PyTorch and OpenCV to streamline image preprocessing and integrate cutting-edge natural language generation (NLG) techniques for accurate text synthesis, enabling high-quality diagnostic reporting.

Engineered a Retrieval-Augmented Generation (RAG) pipeline utilizing FAISS for efficient vector search and LangChain for large language model (LLM) orchestration. Enhanced query precision with semantic vector search, context-aware embeddings, and advanced query refinement strategies, improving document retrieval speed and response accuracy.

Designed and deployed a containerized MLOps pipeline integrated with MLflow to manage model lifecycle. The end-to-end pipeline includes data preprocessing, model training, API deployment (FastAPI), and CI/CD automation. This setup allows for rapid iteration and seamless deployment of scalable machine learning solutions in production environments.

Developed a robust ETL pipeline using Python and pandas for extracting, transforming, and loading large water resource datasets from CSV files. Automated the data preprocessing workflow by handling missing values, standardizing formats, and ensuring high-quality data for subsequent analytical tasks. This solution optimizes the data pipeline architecture for efficient downstream analytics.

Utilized YOLOv5 for object detection in agricultural imagery, focusing on wheat crop head detection. Applied pseudo-labeling and out-of-fold (OOF) validation to enhance model generalization. Preprocessing techniques with OpenCV were implemented, and the model's performance was further optimized with TPU inference, pushing the boundaries of agricultural automation.

Utilized NetworkX and Python for topological analysis of Helsinki's bike network. Visualized traffic flow patterns and identified critical nodes and bottlenecks using Matplotlib, providing actionable insights for urban planners to optimize bike traffic and infrastructure.

Developed a UNet model using TensorFlow for semantic segmentation of salt deposits. Improved segmentation accuracy with data augmentation techniques and fine-tuned loss functions to achieve optimal results for real-world image segmentation tasks.

Performed exploratory data analysis (EDA) and implemented stratified cross-validation with Test-Time Augmentation (TTA) to improve model robustness. Built multi-class classifiers using TensorFlow and Keras, achieving high accuracy for plant disease detection in agricultural images.

Developed a reinforcement learning (RL) agent to play the classic Pong game using Q-learning. Utilized OpenAI Gym for simulation, and TensorFlow for training the agent, applying epsilon-greedy exploration and reward-based learning to enhance agent performance.

Developed a web scraping pipeline using Selenium, Stealth mode, and Chromium for extracting and analyzing flight data. Automated the data parsing and reporting process with Pandas, providing actionable insights for flight price trends.

Leveraged AWS technologies such as SageMaker, Lambda, and CloudWatch to deploy a machine learning pipeline predicting customer responses to marketing campaigns. This system uses historical customer data to improve targeting strategies, optimizing marketing efforts by predicting customer engagement.

Developed a full-stack blog application using Node.js and MongoDB. Integrated Express.js for REST API management and implemented JWT authentication for secure user access. The app supports features like user authentication, blog post creation, and comment sections, with a clean and responsive user interface built with EJS templates.

Developed a full-stack MERN app with JWT authentication for secure user access. Utilized RESTful APIs for backend communication and Redux for state management. Integrated Chart.js for visual data representation and ensured continuous deployment with Docker and automated CI/CD pipelines.

Built a native Android app using Java and XML to facilitate medical Q&A discussions. Integrated a MongoDB backend and RESTful APIs for seamless data communication and user interaction. The platform enables real-time interaction, making medical knowledge accessible to users across various locations.

Developed a real-time location tracking app using Firebase and Google Maps API. Integrated precise geolocation synchronization and route visualization features, allowing users to track real-time positions and view routes on an interactive map, with seamless synchronization across devices.

Created a Flutter-based dashboard for live COVID-19 statistics. Pulled real-time data via REST APIs and displayed dynamic visualizations to provide users with up-to-date information on pandemic trends, while offering an intuitive, interactive UI for better data comprehension.