Salvatore Prochilo

Salvatore Prochilo

AI Engineer · Rome, Italy

Open to relocation within the EU and to remote roles

Download PDF

AI Engineer working on computer vision for Earth Observation. Since 2024 I have built supervised restoration chains for Sentinel-1 SAR and astronomical imagery at Starion, from multi-temporal reference construction to architecture benchmarking and single-pass inference, after three years shipping production software across data, infrastructure and front end. First author and oral presenter at the ESA INSIGHT2026 workshop.

Experience

  1. 10/2024 – present

    AI Engineer

    Starion Group · Frascati (Rome), Italy

    • Authored the denoising section of the state-of-the-art analysis for a national space agency’s on-ground processing chain, then implemented and trained the denoising module delivered for the near-Earth object use case on 16-bit astronomical imagery
    • Built the subpixel co-registration and multi-temporal stacking pipeline that generates pseudo-clean Sentinel-1 GRD references from real acquisitions (five geographic regions, 64 passes each over a three-year span), removing the dependency on synthetic speckle for supervised training Case study: SAR Denoiser
    • Trained a supervised SAR denoiser on real Sentinel-1 noisy–clean pairs: PSNR ≈ 30 dB, SSIM ≈ 0.85, EPI ≈ 0.92 on a held-out Sentinel-1 test set, with single forward-pass inference at deployment instead of 64 acquisitions Case study: SAR Denoiser
    • Benchmarked supervised and self-supervised denoising architectures on Sentinel-1 SAR and 16-bit astronomical imagery, weighing reconstruction quality against inference cost to support architecture selection
    • Assembled an offline document Q&A assistant around BitNet b1.58 2B4T, a natively 1.58-bit model running CPU-only in 0.4 GB of non-embedding memory against 2–5 GB for full-precision peers of the same size, with ChromaDB retrieval, for air-gapped deployment
    • Evaluated the downstream effect of denoising on land-cover classification against a Lee-filter baseline Case study: SAR Denoiser
    • Coordinated the engineering team on a national space agency programme alongside the technical work: acted as the interface between management and the engineers, owned the deliverable documents and kept the team aligned with the delivery schedule

    Technologies: PyTorch, Transformers, OpenCV, GDAL, Rasterio, SentinelHub API, ChromaDB, Weights and Biases

  2. 08/2021 – 09/2024

    Associate Product Software Developer

    Capgemini Engineering · Rome, Italy

    • Engineered the data layer of the grid-monitoring portal of a national electricity transmission operator: parsing a proprietary multi-table file format with Python and Pandas, deduplicating and cleaning records with SQL, and exposing the result through the APIs the front end consumes, improving portal start-up time
    • Unblocked deployment of a telecom vendor’s containerised platform, which dependency conflicts had until then prevented entirely: wrote, with the team, a Kubernetes and Helm module that initialises and bootstraps the microservices, with DaemonSets for per-node components and per-service log visibility boundaries, released through Jenkins CI/CD pipelines
    • Led a small team building an internal reporting portal from scratch, as the most experienced engineer on it: designed the questionnaire workflow and client-side state model in React, and set the technical direction the team followed
    • Developed portlets and reusable components on Liferay for a gaming operator’s retailer portal

    Technologies: Python, Pandas, SQL, React, Docker, Kubernetes, Helm, Jenkins, Git

Education

  • 2023 · Modena, Italy

    Computer Engineering | MSc (Master’s degree)

    Università degli Studi di Modena e Reggio Emilia

    Thesis: Document Image Binarization using Fast Fourier Convolution – AImageLab, advisor Prof. Rita Cucchiara

    Read the thesis case study
  • 2017 · Cosenza, Italy

    Computer Engineering | BSc (Bachelor’s degree)

    Università della Calabria

Skills

Deep Learning & Computer Vision
PyTorch, OpenCV, CNNs, Transformers, image restoration, segmentation, object detection
Earth Observation & Geospatial
Sentinel-1/2, SAR (GRD, speckle, ENL/EPI), multi-temporal stacking, subpixel co-registration (AROSICS), GDAL, Rasterio, SentinelHub API
LLMs & Retrieval
Hugging Face, RAG, semantic retrieval, ChromaDB, quantisation (BitNet)
MLOps & Deployment
Docker, Kubernetes, Helm, Git, Jenkins, experiment tracking (Weights & Biases, TensorBoard), inference profiling, job scheduling on HPC clusters (Slurm)
Programming & Data
Python, Java, SQL, NumPy, scikit-learn, REST APIs, unit testing (pytest)
Ways of Working
technical leadership on a small delivery team, client-facing engagements, conference presentation to a technical audience
Languages
Italian (mother tongue), English (professional working)

Talks & Publications

  • 05/2026 · Frascati, Italy

    INSIGHT2026 – Innovative Signal Exploitation for Geospatial Intelligence

    European Space Agency – ESRIN, Centre for Earth Observation

    From Temporal Stacking to Single-Pass Inference: A Deep Learning Framework for SAR Denoising with Real Sentinel-1 References

    First author, oral presentation

    Read the case study

Projects

Certifications and the full tool list are in the PDF version.