What — Power Asset Forecasting
Description — Time-series analysis. Wind & solar power forecasting mainly for balance market trading.
Developed and implemented the ML life-cycle (data / training / models / inference / mlops) pipeline with the main emphasis on scalable & repeatable methodologies
that could be applied globally with one of the best in market performance and many various sources of input data.
Tools
What — In-flight anomaly detection & localisation on edge AI.
Description — Time-series analysis, anomaly detection, open-set-classification.
Designed and implemented a methodology for anomaly detection in-flight, using Transformers and tree-based models used for live telemetry & diagnosis.
Inferencing in fractions of a second and being fully packaged for an edge device.
What — Satellite based house roof detection and measurements.
Description — Computer Vision, Geospatial Data Analysis. Began with a feasibility study — designed, implemented, optimized for cost and scale data gathering & extraction methodologies
for high-resolution satellite imagery, map data, and 3D models. Then based on the dataset designed and integrated an e2e ML pipeline for hierarhical ML to detect buildings,
their roof sides, and roof-side angles (azimuth, tilt). Later converted into a batch inference workflow using only satellite imagery as inputs.
What — Traffic Sign Detection and Localisation.
Description — Computer Vision (Object Detection, Classification, Object Tracking), Geospatial Data Analysis. Developed and integrated object detection, classification and labeling tools in order to pre-label/predict
traffic signs. In order to reduce costs of manually detecting & localizing & classifing traffic signs.
