AI Engineer (Computer Vision)
IKnowHow SA., Athens, Greece
Nov 2025 – present
- Designed a semi-automated annotation pipeline using YOLO11 for bbox generation and SAM2 for mask refinement via text and bbox-guided prompts, managed dataset versioning on Roboflow
- Developed and benchmarked RF-DETR and YOLO family models for leaf detection ($mAP@50: ~92%$) and tomato ripeness classification on edge devices, authored technical report on accuracy/compute trade-offs
- Implemented a vegetation index-based plant health alert system (NDVI, NDWI, NIR-derived indices) from multispectral imagery to flag disease indicators and estimate water content
- Conducted greenhouse field campaigns using multispectral imaging (RGB + NIR bands) to create a novel dataset for plant disease detection and downstream classification tasks
- Engineered a modular multimodal classification pipeline for RGB and multispectral imagery, improving classification accuracy from $83%$ to up to ~$90%$, enabling flexible experimentation with stacked-band inputs and attention-based multi-branch fusion architectures
- Maintained collaborative ML workflows using Git/GitHub, managed model and dataset versioning via Hugging Face, and performed hyperparameter optimization with Optuna and experiment tracking using MLflow
