om↗Osmar MoralesAI platform engineering
Resume
PAYMENTS EXPERIENCE. FRONTIER MINDSET.Miami, Florida ↗

I make
frontier AI
a team capability.

At Visa, Agent Factory grew organically to 500+ users. I turn new AI tools into shared platforms and useful workflows, so the whole team can move forward.

Explore my work
Osmar Morales

Director, AI Agent Platform · VisaBuilding Elygent as an independent platform project.

⌁ CAPABILITY EXPLORER01 / INTERACTIVE
THE IDEA IN ACTION

One intent.
A governed path.

Choose a workflow. Inspect the decisions behind it.

DESIGN DECISION / Identity & policy

Resolve the caller to the right person or business boundary. A connected AI client gets authorized capabilities, not a shared master credential.

Check scope, permission, and action-specific approval before execution.
An authorized action with a reviewable result.

Conceptual Elygent workflows. Explore the design, not a live production session.

BUILT. ADOPTED. REPEATED.

The platform matters
when people use it.

500+

Agent Factory users

Organic adoption · Sep 2026
600+

Repositories migrated

With 250+ pipelines
Days→hours

Benchmarking turnaround

1–3 days to 1–3 hours

01 / SELECTED WORK

Good AI is useful.
Great platforms multiply it.

Three perspectives on the same problem: making advanced capability work for more people.

VISA · ENTERPRISE AI

Frontier capability.
Everyday adoption.

A platform that makes AI capability accessible across the team.

The challenge
New models and tools arrive faster than people can learn them. Individual experiments need a path into repeatable work.
What I owned
I architect and operate the platform around an existing agent engine, connecting governed tools, reusable skills, knowledge, model routing, and human review.
MCPAgent orchestrationEvaluationPlatform enablement
Read the evidence
OUTCOME VIEW01
500+users · organic adoption
01

Package new capabilities behind a familiar experience.

02

Keep identity, budgets, and approvals at the platform boundary.

03

Use independent validation and feedback to improve delivery.

500+ users, with organic growth. Container, macOS desktop, and private-cloud delivery support different ways of working.

Adoption confirmed September 2026. No daily-active or concurrent-user claim.

02 / MY OPERATING APPROACH

The frontier changes.
The team keeps up.

People shouldn’t have to become experts in every new AI tool to benefit from it. My job is to absorb that learning curve and turn it into shared capability.

Explore how I think
01

Stay at the frontier.

I follow new models, tools, and interaction patterns closely, looking for a concrete problem they can solve for the team.

Frontier learning → shared advantage

03 / THE PLATFORM LAB

Start with the workload.
Make assumptions visible.

A small design exercise. Change the inputs to see why model fit and serving capacity are different engineering questions.

AI capacity sketch

ILLUSTRATIVE
Fictional dense model; parameter count only
Weight precision
Replace this assumption with measured throughput.
WEIGHTS ONLY130.4 GiB

Model parameters × bytes per weight

OUTPUT DEMAND600 tok/s

Requests × output tokens ÷ 60

Weight-fit device floor3
Assumed 80 GiB per device, 20% reserved.
Output-throughput replica floor3
A replica may require multiple devices.

Fit is not capacity. These are lower bounds. KV cache, prefill, activations, p95/p99 latency, and failure redundancy need workload benchmarks.

Design exercise, not production measurements. Memory considerations

04 / THE EXPERIENCE BEHIND IT

Business context.
Engineering depth.
Relentless curiosity.

From analytics to data platforms to agent systems, the throughline is making complex capabilities useful to a whole organization.

Osmar Morales
Osmar MoralesMiami, FloridaLet’s talk
2025 — PRESENT

Director, AI Agent Platform

Visa

Architecture, governance, reusable capabilities, and adoption.
2020 — 2025

Director, Data Engineering

Visa

Data engineering, MLOps modernization, and AI workstreams.
2018 — 2020

Manager, Data Analytics

Visa

Data foundations and self-service pipeline automation.
2016 — 2018

Data Scientist

FedEx Express

Earlier work in revenue systems and process automation.

Education & recognition

BBA, Finance · Mathematics minor
Florida International University, 2015

AA, Mathematics · Miami Dade College, 2008
Harvard Extension School professional development, 2018–2020

Visa LAC Support Excellence Award, 2026
FedEx LAC Professional of the Year, 2017
Five Visa Technical Innovation team awards, 2023–2025

DON’T JUST READ THE RESUME

Have a conversation
with the evidence.

Take a sourced portfolio brief into ChatGPT. Explore the architecture, ask a harder question, or assess the fit.

Uses your ChatGPT session.Download the public evidence