Available now · Remote-first · Immediate start

Manuel Zumbado-Corralesbuilds vision into machines.

Machine Learning Scientist/ Computer Vision/ Deep Learning Research

PhD in Deep Learning with 7+ years across computer vision, medical imaging and large-scale ML systems. I design and train novel models from scratch — like DeepEMSeg for cryo-EM — and bring that depth to foundation models (SAM2, SAM3, ViT) in production.

PhD · Deep Learning IEEE · published author From scratch → production Graduate faculty · TEC
SCROLL
01 / about

Research depth meets production reality

I sit at the intersection of frontier research and shipped systems. For my PhD I designed and trained models from scratch — DeepEMSeg — to automatically segment structural features in cryo-EM protein maps, contributing novel methods for structural biology and drug discovery, published at IEEE conferences with Purdue's Kihara Lab.

That full-lifecycle range carries into industry: I both build models end-to-end and adapt foundation models for production — training custom architectures, deploying SAM2 and SAM3 into annotation pipelines, building CV systems from point-cloud extraction to inference, and driving ML optimization across segmentation, detection and LiDAR 3D projection.

Alongside industry, I teach graduate-level Deep Learning at the Costa Rica Institute of Technology — from training architectures from scratch (CNNs, U-Net, diffusion, GNNs) to modern multimodal systems (ViT, DINOv2, CLIP, Grounding DINO) — and continue research in biomedical imaging and HPC-parallelized image processing.

model_cardv2026.1
namemanuel-zumbado-corrales, phd
taskvision · train + deploy
experience7+ years
locationHeredia, Costa Rica
languagesES native · EN pro
setupremote-first
noticeimmediate
status● open-to-work
02 / selected work

From scratch, to production

research · trained from scratch

DeepEMSeg — cryo-EM map segmentation

My PhD work: designed and trained deep-learning models from scratch to automatically segment structural features in 3D cryo-EM protein density maps — novel methods for structural biology and drug discovery, developed with Purdue's Kihara Lab.

production · foundation models

SAM2 / SAM3 annotation at scale

Deployed Segment Anything foundation models in Databricks for ML-assisted pre-annotation, and built a SAM3 video tracker via Promptable Visual Segmentation — a single bounding box auto-tracks masks across full video sequences.

SAM2/SAM3PVSDatabricksPyTorch
applied DL · geoscience

Seismo-volcanic event classification

Designed and trained deep-learning models to automatically classify seismo-volcanic signals at Turrialba Volcano, in collaboration with OVSICORI — published at IEEE BIP 2023.

03 / experience

A decade shipping & teaching ML

Machine Learning Scientist · Sama
Jul 2025 — Present
San José, Costa Rica · Foundation-model deployment for ML-assisted annotation
  • Lead computer-vision initiatives within the CS team, driving ML integration into annotation workflows for automotive (ADAS), road-sign detection and 2D/3D segmentation.
  • Deployed SAM2 & SAM3 in Databricks as a production PoC for ML-assisted pre-annotation, replacing legacy Voxel51 workflows and running bulk predictions at scale.
  • Built an end-to-end CV pipeline: point-cloud (PCD) extraction → model orchestration → inference → ingestion into the annotation store, validating annotation-ready outputs across the full path.
  • Shipped a SAM3 video tracker using Promptable Visual Segmentation, plus an MVP Chrome extension (TypeScript + Webpack) for off-platform task tracking.
PyTorchSAM2/SAM3OpenPCDetDatabricksMLflowDelta LakeApache SparkAWS S3TypeScript
Teaching Faculty, MSc Computer Science · Costa Rica Institute of Technology
Sep 2019 — Present
Cartago, Costa Rica · Graduate Deep Learning, Machine Learning, Big Data
  • Teach graduate Deep Learning — modern CV architectures (ViT, DINOv2), foundation models (CLIP, BLIP-2, LLaVA, Grounding DINO) and SOTA multimodal systems (Qwen2-VL, GPT-4o, Gemini).
  • Co-design curricula & lab notebooks covering GANs, autoencoders, U-Net, diffusion models and graph neural networks.
  • Delivered courses across Machine Learning, Statistics for Data Science (16 sections), Big Data (13 sections), Data Mining and Business Intelligence.
  • Lectured a graduate course in Data Analysis for Cybersecurity — statistical modeling, anomaly detection and ML for security.
ViTDINOv2CLIPGrounding DINODiffusionGNNsPyTorch
Researcher · TEC / CeNAT
Jul 2018 — Present
Cartago & San José · PARMA Group; National Advanced Computing Collaboratory (CNCA)
  • Designed scalable ETL pipelines with Apache Spark / PySpark for large-scale datasets in S3 warehouses; deployed ML on AWS and GCP.
  • Developed parallel image-processing algorithms for HPC; benchmarked OpenMP, MPI, OpenACC and CUDA across architectures and built reusable data-integration APIs.
  • Built DL models for seismo-volcanic classification at Turrialba Volcano (with OVSICORI); broader research in biomedical image processing within the PARMA Group.
Apache SparkPySparkCUDAOpenMPMPIOpenACCAWSGCP
Visiting Graduate Research Fellow · Purdue University
2020 · 2022
West Lafayette, Indiana, USA · Kihara Bioinformatics Lab
  • Designed processing & curation protocols for cryo-EM map datasets used to train DL segmentation models.
  • Developed Python modules for high-throughput protein structure analysis, cloud integration (AWS, GCP) and structural-descriptor computation on cryo-EM maps.
Pythoncryo-EMstructural biologyAWSGCP
04 / stack

The technical surface

Computer Vision & Model Development // core

Model training from scratchCustom architecture designLoss designData curationU-NetCNNsVision Transformers (ViT)DETRInstance & semantic segmentationObject detectionSAM · SAM2 · SAM3Grounding DINOCLIP · DINOv2PVSPoint clouds (OpenPCDet)LiDAR · 3D projectionBiomedical segmentation

Deep Learning & ML // frameworks

PyTorchOpenCVNumPyPandasscikit-learnMatplotlibMLflowHugging Face TransformersModel deploymentMLOpsETL pipelinesFine-tuning & evaluation

Data & Cloud Infrastructure // scale

DatabricksDelta LakeApache SparkPySparkAWS (S3, EC2)GCPDockerData warehousing

HPC, Parallel & Languages // systems

CUDAOpenMPMPIOpenACCGPU accelerationHPC benchmarkingPythonC / C++TypeScriptBashSQLLaTeX
05 / research

Selected publications

~0
Total citations
Google Scholar
0
h-index
Google Scholar
0
Peer-reviewed works
IEEE · Springer · MDPI
0+
Years of research
since 2017
[01]

Deep Learning Segmentation of Protein EM Maps

Zumbado-Corrales, M. & Esquivel-Rodríguez, J.
IEEE 5th Int. Conf. on BioInspired Processing (BIP), San Carlos, Costa Rica · 2023
DOI ↗ ★ cited by 1
2023
[02]

Automatic Classification of Seismo-Volcanic Signals with Deep Learning: The Case of Turrialba Volcano

Salas, D. A., Zumbado, M., Pacheco, J., Mora, M., van der Laat, L. & Meneses, E.
IEEE 5th Int. Conf. on BioInspired Processing (BIP) · 2023
DOI ↗
2023
[03]

A Comparative Evaluation of Modern Architectures for the Non-Local Means Filter using Performance Primitives Libraries and Compiler Directive APIs

Zumbado-Corrales, M., Castro, J. & Meneses, E.
IEEE 3rd Int. Conf. on BioInspired Processing (BIP) · 2021
★ cited by 1
2021
[04]

EvoSeg: Automated Electron Microscopy Segmentation through Random Forests and Evolutionary Optimization

Zumbado-Corrales, M. & Esquivel-Rodríguez, J.
Biomimetics, 6(2), 37 · MDPI · 2021
★ cited by 7
2021
[05]

Matching of EM Map Segments to Structurally-Relevant Bio-molecular Regions

Zumbado-Corrales, M., Castillo-Valverde, L., Salas-Bonilla, J., Víquez-Murillo, J., Kihara, D. & Esquivel-Rodríguez, J.
High Performance Computing, CARLA 2019 · Springer CCIS vol. 1087 · 2020
2020
[06]

McCulloch-Pitts Artificial Neuron and Rosenblatt's Perceptron: An Abstract Specification in Z

Zamora-Cárdenas, W., Zumbado, M. & Trejos-Zelaya, I.
Technology Inside by CPIC, 5, 16–29 · 2020
★ cited by 6
2020
[07]

Assessing the Impact of the Deceived Non-Local Means Filter as a Preprocessing Stage in a CNN-Based Approach for Age Estimation Using Digital Hand X-Ray Images

Calderón, S., Fallas, F., Zumbado, M., Tyrrell, P. N., Stark, H., Emersic, Z., Meden, B. & Solis, M.
IEEE 25th Int. Conf. on Image Processing (ICIP) · 2018
★ cited by 25
2018
[08]

DNLM-MA-P: A Parallelization of the Deceived Non-Local Means Filter with Moving Average and Symmetric Weighting

Calderón, S., Castro, J. & Zumbado, M.
IEEE Int. Work Conf. on Bioinspired Intelligence (IWOBI) · 2018
★ cited by 5
2018
[09]

DNLM-IIFFT: Implementing the Deceived Non-Local Means Filter Using Integral Images and the FFT for Reduced Computational Cost

Calderón, S. & Zumbado, M.
Progress in Pattern Recognition, Image Analysis, Computer Vision and Applications (CIARP) · 2017
★ cited by 4
2017
06 / education

Foundations & credentials

PhD in Engineering — Deep Learning
Costa Rica Institute of Technology
2020 — 2025
Thesis: automated segmentation of structural features in cryo-EM protein maps using deep learning. Visiting Fellow, Kihara Lab @ Purdue.
MSc in Computer Science
Costa Rica Institute of Technology
2018 — 2020
Licentiate in Computer Engineering
Costa Rica Institute of Technology
2011 — 2017
Additional training: Databricks Machine Learning Fundamentals (certified, 2025) · Deep Learning + Reinforcement Learning Summer School · High Performance Computing Summer School · Big Data Summer School.
Open to senior IC roles

Let's build something that sees.

Looking for senior individual-contributor work in computer vision, multimodal AI, autonomous systems or medical imaging. Remote-first, available immediately.

Senior / Staff ML Engineer Applied Scientist Research Engineer