Thanks for visiting my profile

Dr. Mohammad Amin Khalili

Preferred name: Amin

AI/ML Engineer and Data Scientist with 5+ years of experience developing and deploying scalable machine learning and AI solutions across diverse domains. Experienced in building end-to-end ML systems, from problem definition and data analysis to model development and deployment on AWS and Azure with MLOps and CI/CD practices. Strong background in computer vision, predictive modelling, and data-driven decision-making, delivering measurable improvements in model performance and operational efficiency.

I am an Applied AI/ML Engineer and Data Scientist with a strong track record of building production-ready machine learning systems across computer vision, geospatial analytics, remote sensing, and time-series modelling. I specialise in turning complex data and research-grade methods into scalable, operational AI products that solve real business and industry problems. My work spans machine learning, deep learning, Python, PyTorch, SQL, MLOps, cloud deployment, APIs, and geospatial data engineering.

Currently, I work as a GIS and Remote Sensing Data Scientist, delivering industry-facing AI solutions at the intersection of applied R&D, stakeholder engagement, product delivery, and scalable analytics. I lead projects from problem definition through to deployment, translating business needs into technical requirements, designing robust ML pipelines, validating performance, and integrating outputs into automated production workflows.

My core expertise includes Earth Observation (EO), GIS, remote sensing, SAR/InSAR time-series analysis, computer vision, image segmentation, classification, detection, LiDAR, aerial imagery, satellite data, and multi-sensor geospatial integration. I also build end-to-end ML/DL pipelines using Python, PyTorch, FastAPI, Streamlit, GPU acceleration, reproducible experimentation, and cloud-based AI systems, with a strong focus on scalability, automation, and quality control.

Alongside industry delivery, I have authored 31+ ISI-indexed publications, contributed to editorial activities, and completed 70+ peer reviews across geospatial AI, Earth Observation, and applied machine learning. I am based in the UK and hold a Global Talent Visa.

Leicester, UK · UK Global Talent Visa - No Sponsorship Required · Open to Relocation

LinkedIn Google Scholar Email Me

Work Experience Profile

⚙️

5+ Years

AI/ML engineering & deployment

☁️

AWS & Azure

MLOps, CI/CD, production delivery

🔄

Full Life Cycle

End-to-end ML/DL pipeline ownership

👔

Project Lead

Stakeholder engagement & delivery

Research Profile

📄

31+

ISI-indexed publications

🏛️

14+

International conferences

📋

70+

Peer reviews completed

🏆

Best Paper

EWSHM 2024 Award

Career Journey

Work Experience

GIS & Remote Sensing Data Scientist

Feb 2025 – Present

Bluesky International Ltd (Woolpert) · KTP Associate - University of Leicester & Innovate UK

Led the design, implementation, and deployment of national-scale ML/AI pipelines for tree canopy segmentation and species classification, achieving 80% accuracy improvement and scaling from single-tile to 54,000-tile workloads across the UK.
Developed attention-based ResNet deep learning models achieving 92% classification accuracy; applied NLP and RAG workflows to stakeholder documentation to reduce decision turnaround time.
Owned end-to-end scalable multi-tile workflows using FastAPI, translating business requirements into technical solutions and delivering production-ready AI systems on AWS and Azure with CI/CD pipelines.
Delivered all project milestones ahead of schedule within Agile frameworks, maintaining active stakeholder engagement throughout the full project lifecycle.

Key Projects - Bluesky International Ltd

FastAPI + Streamlit Deployment Platform - LiDAR ITC Segmentation

Designed, implemented, and deployed a production-ready service wrapping the CHM/RGB/LiDAR Individual Tree Crown (ITC) segmentation pipeline behind a FastAPI backend with a Streamlit UI. Implemented job orchestration, health checks, parameterised runs, and live progress polling for both single-tile and batch execution. Automated deliverable generation including QA dashboards, logs, and downloadable ZIP artefacts for stakeholder review and commercial handoff.

FastAPIStreamlitLiDARCHMProduction

Full Life CycleOwnershipDeploymentStakeholder Delivery

Advanced Generalisation & Mixture of Experts (MoE)

Took full ownership of developing a Greedy Ensemble Fusion combined with Domain-Aware Mixture of Experts (MoE) to solve cross-region generalisation challenges. Trained on 110 tiles and tested independently on 8 regions, achieving 85–90% accuracy. Addressed critical operational constraints including domain shift, memory limits, and noisy boundary conditions through tiling, windowed inference, crown-level decision logic, and adaptive expert weighting.

MoEEnsembleTransfer LearningGPU

InitiativeOwnershipR&D LeadershipGeneralisation

LiDAR-Aware Multi-Stage Watershed ITC Segmentation

Independently developed a multi-stage watershed ITC pipeline integrating LiDAR point-cloud structure, CHM, and aerial imagery to strengthen crown boundary delineation in complex canopy conditions. Demonstrated clear accuracy gains in dense forest environments where single-source methods fail. Achieved rapid R&D-to-production translation, commercially deployed on 37 tiles in Iceland following supervisor and client sign-off.

LiDARWatershedComputer VisionCommercial Deployment

Full Life CycleInnovationCommercialisationQC

Production Delivery & Commercialisation, 500-Tile Scale

Led full production delivery of Bankfarm end-to-end outputs (segmentation and species classification), validated through internal QA and confirmed via client commercial approval. Scaled the unsupervised pipeline to 500 tiles (~500 km²), fine-tuning morphological filters for boundary stability and batch robustness. Operationalised C/D inference at national scale, completing 500-tile runs in approximately 3.5 days, with outputs integrated into product workflows.

MLOpsScaleQA/QC500 km²

Project LeadStakeholder EngagementProductionCommercial Approval

Attention-Driven ResUNet, Species Classification (C/D)

Designed and trained an attention-gated ResUNet pipeline to learn crown-relevant spectral and structural features from aligned multi-season RGBI inputs supplemented with CHM and vegetation indices. Used composite loss functions (Dice + Focal + Weighted CE) and produced crown-level classifications through aggregation and calibration. Achieved approximately 0.75 accuracy on independent test regions, demonstrating deployable signal beyond the training domain.

ResUNetAttentionRGBIDeep Learning

Model DesignValidationGeneralisation

Stakeholder Intelligence & NLP Roadmap Prioritisation

Consolidated multi-source stakeholder comments into a single corpus and applied NLP pipelines including weak-label aspect tagging and sentiment scoring (TextBlob polarity) to quantify feedback by product theme. Synthesised actionable product priorities that fed directly into the next NTM requirements questionnaire, reducing decision turnaround time and aligning the technical roadmap with commercial business goals.

NLPRAGTextBlobLLM Workflows

Stakeholder EngagementProduct RoadmapAgileDecision Support

Remote Sensing & AI Specialist / Consultant

Jun 2023 – Dec 2024

SINTEMA Engineering SRL · Italy

Developed AI-based geohazard detection workflows for Sentinel-1 and COSMO-SkyMed SAR imagery, improving detection sensitivity by approximately 40% through advanced ML integration.
Implemented CI/CD pipelines to automate ingestion of new satellite imagery, enabling continuous retraining and improvement of deep learning models in production.
Engaged directly with the Municipality of Naples throughout the full project lifecycle, from solution design and technical guidance through to operational delivery supporting urban planning and hazard decision-making.

Key Projects, SINTEMA Engineering SRL

Stacked Deep Learning for Landslide Reactivation Prediction

Proposed a novel stacked deep learning framework combining GCN, GCN-LSTM, and a meta-model to predict landslide-induced surface deformation. Jointly models spatial geological factor interactions and temporal rainfall-deformation dynamics from COSMO-SkyMed MT-InSAR data. Case study: Randazzo Landslide, NE Sicily. The stacked model significantly outperformed standalone baselines on MAE, RMSE, and R², supporting scalable forecasting for early warning and hazard management systems.

GCN-LSTMInSARTime-SeriesPublished Q1

Research LeadershipPublicationStakeholder Impact

Transformer-Based Models for Landslide Deformation Prediction

Developed and evaluated Transformer-based models for predicting landslide deformations in Caiazzo, Italy, leveraging COSMO-SkyMed imagery and geographical indices (elevation, slope, vegetation). Models analysed temporal sequences from 275 satellite images spanning 2013–2021, capturing critical rainfall–soil deformation relationships through attention mechanisms. Significantly outperformed traditional deep learning approaches. Awarded Best PhD Student Paper at EWSHM 2024.

TransformerCOSMO-SkyMedTemporal AIBest Paper Award

InnovationAward-WinningDeep Learning

SAR Multi-Sensor Fusion with CNN + LSTM for Landslide Monitoring

Developed a synergistic MT-InSAR approach combining C-band and X-band SAR satellites with LSTM and CNN architectures for spatial and temporal interpolation of InSAR deformation results. Applied to active landslides in Cuenca, Ecuador. Achieved average RMSE improvement of 73% at nine validation stations, demonstrating the power of multi-sensor fusion and deep learning for operational geohazard monitoring at scale.

LSTMCNNSAR Fusion73% RMSE gain

Full Life CycleMulti-SensorValidation

Remote Sensing & AI Engineer (Freelance)

Nov 2019 – Nov 2021

Self-Employed

Delivered Earth Observation and InSAR analytics solutions using GMTSAR, StaMPS, and ISCE, streamlining processing workflows and reducing end-to-end turnaround time for clients.
Developed and applied AI/ML models on diverse datasets spanning financial, legal, and geospatial domains, delivering tailored solutions aligned with client requirements.
Managed multiple concurrent projects simultaneously, ensuring on-time delivery and quality outcomes while maintaining client relationships throughout the full engagement lifecycle.

Key Projects, Freelance

InSAR Multi-Hazard Analysis, Earthquake & Subsidence, Mashhad

Applied DInSAR using Sentinel-1A SAR data to simultaneously detect two geophysical hazards in a single interferogram: coseismic displacement from a Mw 6.1 earthquake near the Kashfroud fault, and broad-scale land subsidence across the Mashhad plain attributed to sustained groundwater over-extraction. Demonstrated the unique capability of InSAR for geodetic discovery beyond initial analysis objectives.

DInSARSentinel-1Multi-Hazard

Independent ResearchFull OwnershipDiscovery

Spatiotemporal Subsidence & Change Detection, Tehran Plain

Conducted spatiotemporal characterisation of land subsidence across Tehran plain using InSAR time-series (2016–2021) and Landsat 8 imagery. Integrated NDVI and NDMI indices to bridge satellite observations with hydrological causes. Documented a recent deceleration trend, providing an interpretable evidence base for urban planning and groundwater policy.

InSARLandsat 8NDVI

End-to-End AnalysisMulti-Source DataPolicy Impact

Geospatial & Surveying Engineer

Nov 2015 – Nov 2019

Godakhtar Industrial Group (GIG) · Iran

Conducted GNSS surveying and geospatial data processing for cadastral and large-scale infrastructure projects, delivering precise spatial datasets for engineering and planning applications.

Academic Background

Education

PhD, Remote Sensing, AI & Geosciences

University of Napoli 'Federico II' (UniNa)

January 2022 – December 2024 · PhD Cycle XXXVII

Thesis: Integrated SAR System for Earth Observation and Mitigating Natural and Anthropogenic Phenomena Using Artificial Intelligence Applications

Research Contributions

Developed GCN/GCN-LSTM graph-based deep learning models for landslide deformation monitoring and prediction using SAR/InSAR time-series signals, absolute error below 4 mm for 92% of predictions
Designed unsupervised clustering methodology for APS-corrected InSAR deformation-rate mapping with improved physical interpretability
Pioneered transformer-based modelling for spatio-temporal landslide forecasting, awarded Best PhD Student Paper at EWSHM 2024
Ran complete remote-sensing pipelines: time-series processing, feature engineering, model training/validation, and case-study reporting for multiple applications

MSc, Geodesy, Geomatics & Remote Sensing Engineering

K. N. Toosi University of Technology

September 2016 – February 2019 · Tehran, Iran

Thesis: Deformation analysis of land subsidence using InSAR data and Mesh-free method interpolation

Key Contributions

Applied GRBF interpolation to scattered InSAR vertical deformation data; LGRBF variant achieved +93% accuracy vs FEM and +99% vs polynomial methods on real datasets
Optimised shape parameters via PGRBF and LGRBF (LOOCV); benchmarked against FEM, MLS, and standard GRBF

BSc, Surveying Engineering

Tafresh University

September 2013 – February 2016 · Tafresh, Iran

Foundation in surveying, geodesy, geomatics engineering, GNSS, and micro-geodesy network design.

Activities

Practical internship covering RS, GIS, Micro Station, GPS, and topographic mapping implementation
Proficiency in GNSS survey, micro-geodesy network, and geodetic adjustment techniques

Professional Recognition

Certificates & Courses

🏆

EWSHM 2024 · Potsdam, Germany

Best PhD Student Paper Award

June 2024

Recognised for innovative integration of Transformer architectures with SAR imagery for landslide deformation prediction, demonstrating significant accuracy gains over conventional deep learning benchmarks

🎤

Nottingham Business School · 2025

Speaker, Transforming SMEs with AI 2025

November 2025 · England

Delivered: "From Aerial Pixels to SME Products: AI Tree Segmentation and Species Detection for the National Tree Map", communicating production ML at national scale for commercial EO applications

📡

IEEE GRSS / IADF · 2023

Computer Vision for Earth Observation (1.5 CEU, 15 PDH)

September 2023

Advanced CNN and Transformer architectures applied to satellite and EO data, classification, detection, and segmentation for remote sensing applications

🌐

GATHERS / EU Horizon 2020 · 2024

GATHERS Advanced School (2.0 ECTS, 40h)

12–16 February 2024 · Rome, Italy

Intensive training in InSAR, LiDAR, and GNSS seismology, organised by TU Delft, Sapienza, TU Wien, and UPWr; solved real scientific tasks and defended results

💡

GATHERS Hackathon · 2024

GNSS & SAR Time Series Hackathon (1.0 ECTS, 16h)

17–18 February 2024 · Rome

Implemented and defended GNSS/SAR time-series displacement monitoring applications within competitive international teams

📝

Nature Masterclasses · 2024

Focus on Peer Review (3.5 hours)

3 September 2024

Enhanced peer review proficiency, applied directly in 70+ reviews for Springer Nature, MDPI, and Oxford University Press journals

🏔️

LARAM International School · 2022

Landslide Risk Assessment & Mitigation (46h + field trips)

5–16 September 2022 · Salerno, Italy

Comprehensive landslide risk theory, physics-based and statistical modelling, hazard zoning, monitoring frameworks, and field verification at active slope sites

🗺️

Univ. of Naples Federico II · 2023

Spatial Analysis of Geoscientific Data using GIS (3 CFU)

January–February 2023

Advanced geospatial workflows: raster/vector processing, spatial statistics, and geostatistics, applied to InSAR deformation mapping and hazard characterisation

⛰️

Univ. of Naples Federico II · 2022

Analysis of Slope Processes, Rainfall-Induced Landslides (3 CFU)

February 2022

Hydrological triggering mechanisms, shallow landslide hazard assessment, and slope stability analysis applied to southern Italy case studies

🌋

Univ. of Naples Federico II · 2023

Volcanic Hazard and Risk (3 CFU)

May–June 2023

Volcanic process dynamics, eruption monitoring strategies, and multi-hazard risk frameworks for active volcanic systems in Italy

📊

Univ. of Naples Federico II · 2023

Geostatistics in R (3 CFU)

February 2023

Kriging, variogram modelling, and spatial interpolation using R, applied to characterise geospatial deformation datasets for publications

✏️

Avanti Publishers · Ongoing

Editorial Team, Global Journal of Earth Science & Engineering

Ongoing

Contributing to editorial decisions and manuscript review for international geoscience and engineering submissions

📋

Science Publishing Group · 2023–2025

Editorial Board Member, AJASR

March 2023 – March 2025

Evaluated manuscript scientific quality, methodology, and reproducibility for American Journal of Applied Scientific Research

Dr. Mohammad Amin Khalili

Work Experience

Education

Skills & Technologies

Selected Publications

Certificates & Courses

International Conferences

Get in Touch