Guilherme Albertini

New York City Metropolitan Area

ENGINEER
MACHINE LEARNING
AI INFRASTRUCTURE
DATA SCIENCE
LLM SYSTEMS
OPTIMIZATION
AEROSPACE

The aerospace engineer who now builds the infrastructure that runs machine learning at Meta.

Guilherme Albertini builds large-scale ML infrastructure on Meta's ML Core Foundation team, working across data, modeling, training, and inference. He came to it the long way: the better part of a decade as an aerospace engineer — proving a UCLA CubeSat could survive launch, then writing hypersonics software at Lockheed Martin — before an NYU data-science MS turned him toward machine learning. His most distinctive build is Axle's meeting-allocation engine, grown from a priority-queue heuristic into a provably-optimal CP-SAT solver.

85%Less meeting-allocation time at Axle
3xMore priority-aligned meeting allocations
50%ELFIN CAD simulation-runtime cut
4 yrsELFIN satellites operated in orbit

HAPPENING NOWkept fresh by Aster

STANDING TODAY

SINCE 2025

Four habits carry from qualifying a satellite to running ML at scale.

THE IMPACT

One instinct — model it precisely, then ship the system that solves it — from satellites to ML infrastructure.

EVERYTHING ELSE · 5 WORKS

THE TIMELINE

From qualifying a satellite for launch to building the infrastructure that runs machine learning.

The whole record, newest first – tap any moment and Aster tells the story.

2025–PRESENT

ML infrastructure & inference at Meta

Now on Meta's ML Core Foundation team, Guilherme builds large-scale ML across the full lifecycle he names — data, modeling, training, and inference. It is the layer beneath the applications: the systems that make state-of-the-art models trainable and servable at Meta's scale, and the intended destination of a decade-long move from using ML to building the infrastructure under it.

2024–2025

Production LLM systems in finance

Two back-to-back finance roles turned him into a production-LLM engineer. At Capital One he prototyped a Llama 3.1 conversational agent for Business Cards & Payments and built evaluation pipelines for its semi-structured output; at Axle he shipped a meeting-allocation engine, moving it from a priority-queue heuristic to a CP-SAT OR-Tools solver for guaranteed optimality.

“most recently, I have been tinkering with evaluation pipelines for semi-structured data generated from Llama models”

2021–2024

The deliberate pivot into data science & ML

The deliberate retraining. He left aerospace for an MS in Data Science at NYU, converting engineering intuition into method — Spark ALS recommenders, ranking metrics, a neural-reranking capstone — then spent a freelance year assembling the full applied-LLM stack: data in, model fine-tuned, API out.

2014–2021

Aerospace engineer — satellites, hypersonics, and the first taste of software

The first career. As a UCLA student he proved the ELFIN CubeSat could survive launch — thermal models in Thermal Desktop, Nastran vibration models validated against NASA Wallops shake-table data, and a 50% cut in CAD simulation runtime — then spent nearly three years at Lockheed Martin across manufacturing, hypersonics software, and a first semantic-search tool.

“I then refined two full-model CAD assemblies to decrease simulation runtime by 50%”