Featured Projects
Key achievements
IRGAN: Indoor Radio Map Prediction
State-of-the-art cGAN and Conditional diffusion models for Wi-Fi radio map prediction in indoor environments. Achieved 2.5 dB RMSE accuracy.
Technologies Used
Key Features
- Advanced cGAN architecture for radio map generation
- Diffusion models for high-quality predictions
- Custom loss functions for wireless signal modeling
Multi-spectral Image Fusion
Novel Deep learning architecture for fusing multi-spectral and hyper-spectral satellite images to enhance temporal resolution.
Technologies Used
Key Features
- Novel fusion architecture design
- Temporal resolution enhancement
- Sentinel-2 and Sentinel-3 data processing
French-Bambara Translation Model
Creating the biggest French to Bambara Corpus with a methodoly using Backtranslation, Semi-supervised annotation and uncertainty sampling. Then Fine-tune Mulitlingal models such NLLB-200 600M.
Technologies Used
Key Features
- Transformer sequence-to-sequence fine-tuning
- Custom parallel corpus creation
- Data augmentation techniques
Diffusion Models for Mobility Trajectory Generation
Developed Diffusion models (DDPM) to predict Synthetic mobility trajectories from pure gaussian noise.
Technologies Used
Key Features
- A UNet like architecture designed for 1D sequences
- Test of different Noise scheduler (linear, cosine)
- Evaluation of the utility
Automatic Building Plan Vectorization
Supervised Master's thesis project on converting 2D building plan images to vector format using deep learning techniques.
Technologies Used
Key Features
- Deep learning-based segmentation and Novel centerline detection method
- Harris corner detection to detect POI
- Automated vectorization pipeline
Setup a Private Overleaf
I setup a private Overleaf instance for my research group in OBS for privacy and data sovereignty. Used Docker containerization to deploy a full-featured collaborative LaTeX editing server with real-time synchronization.
Technologies Used
Key Features
- Self-hosted collaborative LaTeX editor with real-time multi-user editing
- Secure document storage with complete data privacy for sensitive research
- Docker-based deployment for easy maintenance and scalability
Airflow Migration to Docker
Proposed and implemented a Docker-based solution for Apache Airflow, transitioning the team from a bare metal installation to a containerized architecture using Docker Compose for improved maintainability and resilience.
Technologies Used
Key Features
- Containerized Airflow deployment for simplified maintenance and updates
- Improved system resilience with isolated services
- Standardized development environment
Interested in Collaboration?
I'm always open to discussing new projects, research opportunities, or technical challenges.