GoogleGenAI Forward Deployed Engineer

Sebastian Mateus

I build and break intelligent systems.

I embed with Google's enterprise customers to take agentic AI from prototype to production. Behind that: 8+ years shipping ML systems, and independent research in agent security, LLM evaluation, and mathematical reasoning.

8+Years shipping production ML
917Kaggle global rank, of 224,244
28Competitions · 5 medals · 3 running now
13Engineers on the team I founded

NOW · SINCE NOVEMBER 2026

GenAI Forward Deployed Engineer at Google

An embedded builder role. I work inside strategic customers' environments, closing the distance between frontier AI and production reality.

01

Build inside the customer

Coding, debugging, and shipping bespoke agentic systems in the customer's own environment, rather than advising from the outside.

02

From demo to production

Working the hard middle: integration complexity, data readiness, state management, evaluation, and reliability.

03

From the field to the roadmap

Turning what breaks in real deployments into input for the Google GenAI product roadmap.

WHOAMI

Engineer by trade,
researcher by trajectory.

I spent eight years learning how ML systems fail in production (drift, silent data bugs, ungrounded generations) and building the pipelines, evals, and observability that catch it. That same discipline now drives my research.

Also: TEDx speaker on AI, international hackathon winner (Baubap ML Hackathon, 1st place), trilingual (ES · EN · PT), and mentor to 13+ engineers.

about.py
class SebastianMateus:
    role      = "GenAI Forward Deployed Engineer"
    org       = "Google"
    base      = "Bucaramanga, CO · remote (UTC-5)"
    research  = ["agent security", "LLM evaluation",
                 "mathematical reasoning"]
    stack     = ["Python", "PyTorch", "LangGraph", "SQL",
                 "dbt", "BigQuery", "Docker", "Terraform"]
    oss       = ["eml-translator", "barrio-mapper",
                 "realh"]
    languages = ["es", "en", "pt"]
    kaggle    = "Expert · rank 917 / 224244"
    ships     = True

RESEARCH

What I'm investigating

Personal research, outside my day job, run like production engineering: a public artifact and a reproducible pipeline behind every line of work.

AGENT SECURITY · OPENAI × GOOGLE × IEEESILVER · TOP 3%

Multi-step tool attacks on LLM agents

Algorithmic red teaming: attack algorithms that surface reproducible multi-step failures in tool-using agents, under public and private guardrails. Silver medal, 131 of 4,186 teams. The lever was evaluation discipline: a harness that estimated true scores without spending scarce submissions.

Competition ↗
3D SEGMENTATION · VESUVIUS CHALLENGEWRITE-UP

Reading a scroll that cannot be unrolled

Detecting papyrus surfaces inside CT scans of carbonized Herculaneum scrolls. I designed a 3D hysteresis fusion: high-confidence voxels seed a morphological propagation inside a two-model mask, and its private score beat its public score.

Competition ↗ Write-up ↗
MATHEMATICAL REASONING · ARXIV:2603.21852OPEN SOURCE

Single-primitive math for AI: the EML operator

A Python library for the exp-minus-log primitive, the NAND gate of continuous mathematics, released as eml-translator. It translates expressions to and from EML trees, with 116 tests across Python 3.9 to 3.13 and a benchmark of the operations as LLM agent tools.

Library ↗
SCIENTIFIC ML · IMAGE FORENSICSTOP 8%

Detecting forgery in biomedical research images

Copy-move forgery detection and segmentation in scientific figures, a direct attack on research fraud. Bronze medal, 128th of 1,564 teams, building on a public baseline by abhishekgodara and weights by pankajiitr, rebuilt around a different segmentation head.

Competition ↗ Write-up ↗

CAREER

Professional experience

Eight years, from analytics to production ML to Google.

Nov 2026 – Present

Google · Colombia

GenAI Forward Deployed Engineer (FDE)

Embedded with strategic Google customers, taking agentic systems from prototype to production.

Aug 2026 – Oct 2026

Evolv Consulting

Senior AI Engineer

Nov 2025 – Aug 2026

Nequi (Grupo Bancolombia) · Colombia

Senior AI/ML Engineer

Built a foundation model from scratch, pre-trained on customer transaction sequences, and ran the data and deployment infrastructure behind credit risk models on AWS and Databricks.

Jun 2024 – Jan 2026

Factored · Palo Alto, remote

Competitive ML Team Lead & Founder · ML Engineer

Founded the competitive ML practice and grew it from 1 to 13 engineers in ten months, while shipping RAG, LLM evaluation, and data platforms for client products.

Jul 2023 – Apr 2024

Intouch CX · Winnipeg, remote

Senior Data Scientist

NLP and demand forecasting behind workforce planning.

2018 – 2023

Earlier roles · Robin Food, Immersion Design, Vision Puntual (co-founder), Metric Analytics

Analytics, NLP, pricing, and computer vision across startups, including a data science practice I co-founded.

COMPETITIVE ML

The leaderboard record

Competition is my favorite forcing function: a frozen metric, a hard deadline, and no partial credit for good intentions. Twenty-eight entered, five medals.

Selected write-ups

Technical write-ups of approaches I built for these competitions.

  • NVIDIA Nemotron ↗ Merged several LoRA adapters into a single rank-32 adapter with truncated SVD, the Eckart–Young optimal low-rank approximation.
  • CAFA 6 ↗ Built the Gene Ontology DAG and applied topological max-propagation so predictions satisfy the true-path rule by construction.
  • NeuroGolf 2026 ↗ 400 hand-optimized per-task ONNX graphs, scoring 4,178 → 7,277 over three months of iteration.
  • Santa 2025 ↗ STRtree-indexed parallel packing search, assembled from three credited public ideas into a faster combined solver.

CONTACT

Let's talk about hard problems.

If you're building (or breaking) intelligent systems and want to compare notes, the fastest way to reach me is LinkedIn.