Independent research lab · Querétaro, México

We build AI that
passes clinical validation,
not just demos.

The Medellin AI Lab works on trustworthy machine learning for medical imaging, clinical agents, and the messy problems where models meet real patients. Move your cursor over the field ↑

See research lines Request collaboration
↓

// 01 — Mission

Most medical AI never leaves the slide deck. We care about the part that survives contact with a clinic — the validation, the edge cases, the data you can actually trust.

What we are

A small, independent lab. Focused research on AI for medicine, shipped and written up in the open.

How we work

Reproducible pipelines, honest baselines, clinical relevance over leaderboard scores.

Why "Medellin"

It's the founder's surname — the academic tradition of naming a lab after its principal investigator. The Topol Lab, the Zou Lab… the Medellin AI Lab.

// 02 — Research lines

Where we point the lens.

01

Medical Imaging

Segmentation & classification across modalities — models that hold up across scanners and sites.

  • Segmentation
  • Ultrasound
  • Retina
  • MRI
02

Clinical AI Agents

Agents that follow patients between visits and escalate when it matters. Home of Cualli AI.

  • LLM agents
  • Triage
  • Spanish NLP
03

Robustness & Domain Shift

Domain adaptation so a model trained in one hospital holds up in the next.

  • Domain adaptation
  • Generalization
04

Validation & Trust

Evaluation that mirrors clinical reality: calibration, failure analysis, regulator-grade rigor.

  • Calibration
  • Eval
  • Reg-ready

// 03 — Selected work

Track record.

2026

Cualli AI — clinical follow-up agent

Functional MVP: a WhatsApp-native dual-persona agent that follows diabetic patients and escalates to their doctor. Live system, not a mockup.

Product · MVP shipped
2026

IEEE FETUS Challenge

Participated — AI for fetal ultrasound under real acquisition variability.

Competition · Participant
2026

Foot-ulcer segmentation under domain shift (FUSeg)

Domain adaptation for diabetic foot-ulcer segmentation across imaging sources.

Research
2026

Speaker — IEEE ISBI

Talk on semi-supervised learning and cross-system ultrasound.

Conference · Talk
2025

Speaker — Pediatric Academic Societies (PAS)

AI-powered single-view portable pediatric echo to detect congenital heart disease.

Conference · Talk
Rodrigo speaking at Pediatric Academic Societies 2025
PAS 2025 · pediatric echo AI talk
Rodrigo at Mayo Clinic Platform
Mayo Clinic Platform · 2024–25
Hackathon Conciencia 2026 team
Hackathon Conciencia 2026
Cualli AI patient report on a phone

// Spotlight

Cualli AI

Clinical follow-up agent

Lives on WhatsApp and follows patients between visits — a dual-persona agent that talks to the patient, briefs the doctor, and escalates on its own. Live MVP, real data.

Visit Cualli AI ↗
VíaVital safe-route engine over Mexico City

// Spotlight

VíaVital .ai

Safe-route agent · Hackathon Conciencia 2026

Routes that dodge crime and crashes, not just traffic. An agent reads open CDMX accident & robbery data and traces the safest path by travel mode — pedestrian, bike, car. No invented numbers.

Live demo ↗ GitHub ↗

// 04 — News & notes

What we're publishing.

Lab note

The lab is live.

Setting up The Medellin AI Lab as the home for our medical-AI research. Datasets, write-ups and results will land here.

Project

VíaVital.ai — safe routing

Risk-aware routes over open CDMX data. Demo live.

Coming soon

Open datasets & baselines

Reproducible baselines for our imaging work.

More notes as we ship. Want updates? Reach out.

Rodrigo Medellin Robles presenting at IEEE ISBI Presenting at IEEE ISBI

// 05 — Founder

Rodrigo Medellin Robles

AI Scientist · Principal Investigator

AI scientist working in clinical medtech, with an MSc in Artificial Intelligence. First place in the IEEE FETUS 2026 challenge, ISBI speaker, and founder of Cualli AI. I build clinical AI that passes validation — then write down how.

rodrigo@medellin-lab.com Google Scholar ↗ GitHub ↗

// 06 — Contact

Datasets, collaborations,
challenges, press.

For dataset access requests, research collaborations or anything else — write directly. We reply.

or just email rodrigo@medellin-lab.com