PORTFOLIO // 2026 AVAILABLE · VALENCIA, ES

Juan Felipe
Jaramillo·Hernández

ML / Software / Electronics Engineer · Computer Vision Researcher

I build vision systems that see depth — from YOLO-based detectors on microcontrollers to the anomaly-detection pipeline behind Aimira. PhD candidate at UPV, inventor of the Depth Object Detector.

View work → Request CV

PhD
UPV · CS
MSc
AI · UPV
Lead
Aimira CV

PERSON · 0.99

VISION LEAD · 0.97

INPUT · 1920×1080 · 30FPS

model: DOD-nano · conf≥0.25

0.4mDEPTH∞

    COMPUTER VISION/DEPTH ESTIMATION/YOLO ARCHITECTURES/EDGE AI/PYTORCH/IoT & EMBEDDED/PRECISION FARMING/LIGHT FIELDS/
    COMPUTER VISION/DEPTH ESTIMATION/YOLO ARCHITECTURES/EDGE AI/PYTORCH/IoT & EMBEDDED/PRECISION FARMING/LIGHT FIELDS/
  

61.46%

DOD size reduction

1.6 MB

on-device model

1.0M

parameters

peer-reviewed papers

[ 01 ]  Featured work

Depth Object Detector — a detector that also regresses metric depth.

DOD inference — apple detection with regressed depth

DOD inference

MinneApple · box + score + depth

A lightweight, state-of-the-art convolutional detector (~1M params) that integrates depth estimation directly inside the YOLO prediction heads. Each detection tensor carries an extra output that regresses depth jointly with the bounding box and objectness — derived end-to-end through an augmented objective function. Built for fruit-harvesting and phenotyping on low-cost, resource-constrained edge devices.

PyTorch YOLOv8 COCO + MinneApple Edge / IoT

4.18→1.6

MB, optimized

~1.0M

parameters

+depth

per detection

Repository → Read the paper

[ 02 ]  Open source

Repositories

● querying api.github.com…

LIVE · github.com/Jaramilloh

CURATED · cached fallback

all →

{{ repo.name }} ★ {{ repo.starsLabel }}

[ 03 ]  Research

Publications

ORCID 0000-0002-0992-1346 · VALGRAI / Universitat Politècnica de València

2024

Sensors (MDPI) · Journal

Application of Machine Vision Techniques in Low-Cost Devices to Improve Efficiency in Precision Farming

Jaramillo-Hernández, Julian, Rincón, Marco-Detchart — conceives and benchmarks the DOD for object detection with integrated depth estimation against state-of-the-art models.

DOI ↗

2024

PAAMS · Springer · Chapter

Efficient Depth Object Detection: Ablation-Driven Optimization for Lightweight YOLOv8 Architecture

An ablation study streamlines the DOD by ≈61.46% (4.18 MB → 1.6 MB) while preserving precision on MinneApple — fit for microcontrollers in edge computing.

DOI ↗

[ 04 ]  Capabilities

Stack

Vision & ML

Object detection · YOLODepth estimationPyTorch · OpenCVModel optimization · ONNXLight-field / plenoptic

Languages

PythonC / C++JavaScript / TSMATLABSQL

Edge & Systems

Microcontrollers · IoTEmbedded visionElectronics designCameras & sensorsReal-time pipelines

MLOps & Dev

CI/CD · IaCDockerBackend + FrontendGit workflowsCloud deployment

[ 05 ]  Trajectory

Path

PRESENT

Artificial Vision Lead — Aimira

Leading the computer-vision solution for anomaly detection · Spain. Pure development across backend, frontend, ML and electronics.

2024 — PRESENT

PhD Candidate, Computer Science — UPV

Resource-efficient computer vision: detection, depth and edge deployment for the IoT.

2024

DOD published — Sensors & PAAMS

Invented and benchmarked the Depth Object Detector; ablation-optimized to 1.6 MB.

2022 — 2024

MSc — AI, Pattern Recognition & Digital Imaging, UPV

Valencian Graduate School & Research Network of Artificial Intelligence (VALGRAI).

FOUNDATIONS

BSc Electronic Engineering — P. U. Javeriana

Bogotá, Colombia · early work on light fields & neural image sub-sampling.

* dates are editable — adjust any entry to your exact record.

[ 06 ]  Off the clock

Album

LIVE · ALBUM · {{ photoCount }}

MEDIASOURCE · AWAITING

PHOTO OF THE MONTH

PHOTO OF THE MONTH

StaticManifestMediaAdapter No photos yet. Populate assets/album/photos.json with your image entries — the grid fills from your committed manifest automatically, nothing else changes.

[ 07 ]  // REACH_ME

Let's build something

One ping away — open to research collaboration, vision / edge-AI consulting, and engineering roles. Pick a channel, or take the whole CV.

↗ View CV

live anchors · GitHub, LinkedIn, ORCID & Google Scholar

[ 08 ]  Notes

Writing

module: ContentModule — plug a Markdown / CMS adapter to publish.

Guía · ES/EN

Armá tu entorno de desarrollo con IA — Windows + WSL2

publicado · ~30 min · 9 etapas

Nota · ES/EN

Shrinking a detector 61% with ablation-driven design

publicado · ~6 min · 6 secciones

Nota · ES/EN

From Lytro raw to sub-aperture light-field stacks

publicado · ~10 min · 6 secciones

Nota · ES/EN

3D figures in three.js — robotic arm & FENOBOT

publicado · 6 demos · three.js / WebGL

[ 10 ]  // RECOMMENDED_READS

Papers I'd hand you

not mine — the work that shaped how I think about vision & language

5 entries · curated · class ∈ {CV, NLP}

        ClassPaperYearSource
      

ImageNet Classification with Deep Convolutional Neural Networks

Krizhevsky, Sutskever, Hinton · NeurIPS — lit the deep-learning fuse; CNNs crush ImageNet.

2012 NeurIPS ↗ CV

Deep Residual Learning for Image Recognition

He, Zhang, Ren, Sun · CVPR — skip connections make 100+ layer nets trainable.

2015 arXiv ↗ CV

You Only Look Once: Unified, Real-Time Object Detection

Redmon, Divvala, Girshick, Farhadi · CVPR — detection as a single regression pass — real-time.

2016 arXiv ↗ NLP

Attention Is All You Need

Vaswani et al. · NeurIPS — the Transformer — attention replaces recurrence.

2017 arXiv ↗ NLP

BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding

Devlin, Chang, Lee, Toutanova · NAACL — bidirectional pretraining redraws the NLP baseline.

2019 arXiv ↗

placeholder list — swap in your own picks any time