Overview
During my role as a Software Developer Intern (June–September 2024), we developed an AI-powered nowcasting model to predict thunderstorm formation using radar data. The project involved building a full machine learning pipeline, optimizing model performance, and deploying a lightweight web-based interface.
📄 Initial Research Paper (PDF):
Download Draft
Draft Notice: This paper reflects early-stage work.
The most recent version is not available here.
🔧 Data Engineering
- Processed 20GB+ of raw radar data into high-resolution image tensors.
- Built a data transformation pipeline supporting augmentation, normalization, and sliding-window sampling.
- Automated preprocessing scripts using Python, NumPy, and OpenCV.
🤖 Model Development
- Designed and trained deep learning models using TensorFlow and PyTorch.
- Achieved 92% prediction accuracy by tuning architectures and hyperparameters.
- Reduced model inference runtime by 50% through optimization and pruning.
🌐 Full-Stack Deployment
- Created a Flask backend to serve predictions via REST API.
- Built a clean frontend using HTML, CSS, and JavaScript for radar-image upload and inference display.
- Deployed the system on a lightweight local server for live demonstration.
Skills & Technologies
Python, TensorFlow, PyTorch, NumPy, OpenCV, Flask, JavaScript, HTML/CSS, Data Pipelines, Model Optimization, Radar Data Processing
Result
The project delivered a robust and efficient nowcasting prototype capable of predicting thunderstorm development from radar snapshots, demonstrating both machine learning and full-stack engineering skills.