Hi, I'm João Ataíde.

Projects

Projects and open content

Experience

Professional journey

1. May 2026 – present

   Data Scientist

   Imagem Geosistemas

   • I design and technically lead data, GeoAI and computer vision solutions, from problem definition to architecture, validation and integration into production.
   • I provide technical leadership on projects and consulting engagements: architecture, methodology, schedules, validation criteria, technical risks and team direction across the whole development cycle.
   • I build multi-sensor pipelines combining optical and SAR imagery, vector data, weather data, field cameras, IoT sensors and operational records, using the ArcGIS ecosystem to process, analyze and publish the results.
   • Built a pipeline for anomaly detection in Sentinel-1 SAR time series with TCN-v2, combining Sentinel-2 and optical change detection to identify changes and generate georeferenced alerts.
   • Built 10 m land use and land cover classification with Sentinel-2 and XGBoost, plus an AHP (Analytic Hierarchy Process) multi-criteria model combining NDVI/EVI, proximity, territorial context and operational variables to prioritize critical areas.
   • I am designing a water security solution with Sentinel-1/2 and a digital terrain model (DTM): water body classification, level and volume estimation, temporal analysis and integration with ArcGIS Enterprise/Image Server.
2. Apr 2026 – Jul 2026

   Data Science and Computer Vision Consultant

   Automaxia

   • Built computer vision solutions for automated analysis of biomedical images, with detection and segmentation models to identify and quantify structures.
   • Compared YOLO and RF-DETR using image slicing, prediction fusion and systematic metric and error analysis to select the most suitable approach.
   • Designed ML/DL experiments covering data preparation, validation and comparison of approaches, documenting results, limitations and technical recommendations.
3. Apr 2024 – May 2026

   Data Scientist

   VEGA Monitoramento

   • Built multi-temporal pipelines in Python, Spark and Databricks, integrated with S3 and STAC, for acquisition, mosaics, cloud masks, temporal composites, spectral indices and processing of Sentinel-1/2, Landsat and MODIS.
   • Built an experimentation and training framework with U-Net, DeepLab, Mask R-CNN, MLP and LSTM, integrating Scikit-learn, TensorFlow/Keras, PyTorch, PyCaret, Optuna, cross-validation and experiment tracking.
   • Developed Random Forest and neural network models for land use and land cover, native vegetation and deforestation at national and international scale (Brazilian biomes and countries in South and North America), with feature engineering, time series, class and geographic balancing, outlier handling and change detection.
   • Implemented workflows with SAM, statistical cubes, NDVI/EVI and SAR processing, automating ingestion, geospatial enrichment, quality controls and product generation in a distributed environment.
4. Jan 2024 – present

   Graduate Program Professor

   ClickGeo Cursos e Pós-Graduações em Geotecnologias

   • I teach hands-on courses on Python and R applied to Geographic Data Science.
5. Oct 2020 – Apr 2024

   Data Scientist / Computer Vision Engineer

   Imagem Geosistemas

   • Built multi-sensor, multi-resolution solutions with CBERS-4A, Sentinel-1/2, ICEYE, Landsat, night-time imagery, high-resolution imagery and drones for environmental, urban, agricultural and infrastructure applications.
   • Developed detection, segmentation and classification models for rooftops, pools, green areas, public squares, solar panels, firebreaks, infrastructure, roads, paving and land use and land cover (with imagery down to 50 cm), reaching 98% for street lights, 97% for illegal mining and pools, 85% for radar-based deforestation and over 93% for rooftops in regular areas (85% in complex areas).
   • Built CycleGAN solutions for SAR image colorization and U-Net for land use and land cover, plus change detection and continuous monitoring with optical and SAR data and automated alert generation.
   • Built and operated end-to-end pipelines to detect unauthorized connections, combining computer vision and tabular data, delivering operational gains and revenue recovery.
   • Designed the initial architecture and proofs of concept for GenAI applications with LLMs and geospatial data, built optimization models for crew scheduling, and contributed to a task orchestration and allocation platform.