Computer Vision Researcher

Frontier's Execution Intelligence platform processes real-world imagery to produce structured, decision-ready output — from aerial and satellite data through to clean vector geometry that operations teams can act on. Our Sofia research team owns the computer vision work that powers this pipeline.

The Role

You will join our AI research team in Sofia as a computer vision specialist. Our vision work turns imagery into structured, decision-ready output — segmenting and detecting objects, and extracting clean vector geometry such as building and roof polygons from aerial and satellite imagery.

You own the research side of these pipelines: curating datasets, designing and training models, debugging the training process, and proving out architectures before they reach production. You partner with AI engineers to hand off models in a productionisable state, and you set the standard for evaluation rigor on vision tasks.

Tech Stack

• Core: Python, PyTorch
• Vision: Segmentation and detection frameworks, OpenCV
• Structured prediction: Graph neural network libraries (PyTorch Geometric, DGL)
• Geospatial: GDAL, rasterio, and related remote-sensing tooling
• Supporting: Hugging Face, Databricks, experiment tracking

What We're Looking For

• Research background — MSc or PhD with a computer vision focus, or equivalent demonstrated research experience.
• Core computer vision — Image segmentation, object detection, and/or polygon and vector extraction from imagery.
• Deep learning depth — You train, debug, and evaluate vision models from scratch, not only call pretrained APIs.
• Graph neural networks — A strong advantage, including techniques for structured prediction and optimal-transport matching (e.g., Sinkhorn) for tasks like vector/polygon extraction.
• Geospatial / remote-sensing imagery — Experience working with aerial or satellite data is a plus.
• From paper to prototype — You can implement and adapt recent CV papers in code.
• Dataset and evaluation discipline — Curation, metrics, and honest reporting of model quality.
• Communication — Clear written research and findings for a distributed team.