Solo-Built Agentic LLM Platform Leadership

FABIÁN ROMERO

Senior Software Engineer (AI) — Agentic Platforms & AI-Enabled Delivery

@ ********@*****.*** +525********* # C. Varsovia 36, Juárez, Cuauhtémoc, 06600 * Mexico City, Mexico (Remote)

@fromeroj ï https://www.linkedin.com/in/fabian-romero-11b356b/ § fromeroj PROFESSIONAL PROFILE

Hands-on technical leader with 25+ years building scalable, secure production systems — and the last year spent do- ing precisely what this role describes: leading design and delivery of agentic applications from concept through gov- erned production. Built and operated, solo, an LLM agent platform over a 6,400+ file enterprise estate: orchestration with tool-calling patterns, retrieval and context-aware workflows, planning loops with memory and grounding, human- in-the-loop sign-off boundaries, evaluation gates in CI, monitoring, guardrails, resiliency, cost control and rollback — an estimated 18-month program delivered in 6. Champion of the AI-enabled SDLC in practice: AI-driven test generation, code review quality and release confidence, used responsibly across the lifecycle. Deep Java (nine years of Spring plat- forms serving 100,000+ users) and current Python (FastAPI), with React/Next.js on the frontend; cloud-native on Ku- bernetes across AWS, GCP and Azure; DevSecOps and CI/CD as habit. Security background suited to AI governance: formal security audits, zero-trust (mTLS), encryption, secrets management — reinforced by formal-methods research (4 international publications). Builder of reusable platforms a 52-engineer organization ran on; mentor and design reviewer throughout.

TECHNICAL SKILLS

Agentic AI in Production

• Design: orchestration, tool-calling patterns, planning and rea- soning loops, memory, grounding, retrieval and context-aware workflows

• Safety: human-in-the-loop controls, staged guardrails, unattended- vs-sign-off boundaries, failure handling and dead-letter triage

• Serving: self-hosted open-weight LLMs (vLLM), GPU scheduling; LangChain-class frameworks, RAG, vector search

Engineering Patterns for Production AI

• Evaluation: eval gates in CI, regression on model/prompt change, drift caught before users; production feedback loops

• Operations: monitoring, resiliency, cost control (per-stage cost/latency/token throughput), rollback strategies, versioned prompts

• Governance: concept experimentation governed produc- tion; audit-grade logging; responsible-AI discipline AI-Enabled SDLC

• Delivery: AI-driven design, coding, test generation, documenta- tion and debugging — with engineering judgment kept firmly in charge

• Quality: code review quality, release confidence, automated unit/integration/e2e gates wired into CI/CD

• Productivity: developer workflows and automation; measurable outcomes (an 18-month program delivered in 6, solo) Java & Python (Core)

• Java: nine years of Spring platforms serving 100,000+ users — deep, not historical

• Python: FastAPI, Django, Pytest, asyncio; Pandas, NumPy — cur- rent daily language

• Frontend: React, Next.js, TypeScript/Node.js — intuitive product experiences included

Cloud-Native & Platform Engineering

OTHER ACCOLADES

Bronze medal, International Mathemat-

ical Olympiad (IMO)

Silver medal, Olympiade mathématique

Asie-Pacifique (APMO)

1st Place National Contest of Science

and Technology (Electronics)

LANGUAGES

Spanish (L)

English (C1)

French (B1)

EDUCATION

Master of Science in CS

University of Mexico, UNAM

z 2005-2007

Bachelor of Science in CS

University of Mexico, UNAM

z 2001-2005

PUBLICATIONS

Axiomatic systems and topological semantics for

Intuitionistic temporal logic

JELIA 2019

Decision procedures for epistemic logic exploiting belief bases

AAMAS 2019

• Cloud: AWS (EC2, ECS, EKS, Lambda, S3, RDS), GCP (Cloud Run, Pub/Sub, BigQuery), Azure; managed services and hybrid

• Platform: Kubernetes, Helm, Docker; Terraform IaC; event-driven architectures, data pipelines, integration patterns

• CI/CD: GitHub Actions, GitLab CI, ArgoCD, Flux; DevSecOps practice; observability with Prometheus, Grafana, OpenTelemetry Security for AI Systems

• Controls: authentication/authorization (OAuth 2.0, JWT, RBAC), secrets management (Vault), encryption at rest & in transit, se- cure deployment

• Depth: formal security audits, zero-trust (mTLS), threat-informed design; protocol verification research (4 publications)

• Regulated: 99% availability SLA in regulated fintech; compliance- grade infrastructure discipline

Leadership & Platform Building

• Reusable platforms: frameworks and internal capabilities a 52- engineer organization ran on (Flint)

• Technical leadership: architectural decisions, design and code reviews, hands-on contribution, mentoring across teams

• Stakeholders: roadmaps, risk and trade-off communication to senior leadership; product mindset from 25 years of delivery SAT for epistemic logic using belief bases

EMAS 2019

Decidability of Deterministic Process Equivalence

for Finitary Deduction Systems

PDP 2020

PROFESSIONAL EXPERIENCE

199*-****-**** 2020

Dive Consulting (MX), CTO

Independent Consultant

Wrapit PLC (UK), Lead Architect

Flint Technologies (UK), Technical Director

Pearl Traveltech (US), Technical Lead

Université Paul Sabatier (FR), Researcher

Robeau (FR), IoT Architect

Broxel (MX), Blockchain Architect

MIM (MX), Principal Consultant

Principal AI & Architecture Consultant

Materiales Industriales de Mexico

z June 2025 – Present * Mexico City, Mexico

• Led hands-on design and delivery of an agentic application from concept through governed production, solo: LLM or- chestration with tool-calling, retrieval and context-aware workflows over a 6,400+ file enterprise estate — with eval- uation gates in CI, guardrails, human-in-the-loop sign-off boundaries, monitoring, cost control and rollback. An esti- mated 18-month program delivered in 6.

• Championed the AI-enabled SDLC in practice: AI-driven test generation, code review quality and release confidence, with engineering judgment owning every merge.

• Delivered the governed production platform: 250+ versioned services, PostgreSQL (migrated from SQL Server), Docker/Linux deployment, CI/CD quality gates, zero data egress.

• § github.com/Rurbina1785/MIMV2

Technology stack:

Agentic Orchestration, tool-calling, retrieval, human-in-the-loop, guardrails, evals Discipline Monitoring, cost control, rollback, versioned prompts, drift regression Platform 250+ services, PostgreSQL, Docker, Linux, CI/CD Principal Blockchain Architect

Broxel

z January 2025 – December 2025 * Mexico City, Mexico

• Architected and deployed a production distributed ledger registering real-money transactions — consensus protocol, cryptographic security, encrypted communication — sustained at a 99% availability SLA in a regulated financial envi- ronment.

• Owned security end-to-end: formal security audits, zero-knowledge proof integration, scalability and resilience load- testing before launch — governance-grade delivery discipline.

• Built containerized infrastructure (Docker) with monitoring and observability for production readiness. Technology stack:

Security Formal security audit, zero-knowledge proofs, encryption Distributed High-availability ledger for real-money transactions Infra Docker, CI/CD, monitoring, observability

IoT Architect & Technical Lead

Robeau

z May 2021 – December 2024 * Paris, France

• Architected Go microservices and event-driven pipelines (Kafka) ingesting 10,000+ real-time metrics on Kubernetes/Docker/G

— highly available cloud-native services with strong observability, built, run and maintained for three years.

• Kept production ML healthy end-to-end: forecasting and anomaly-detection models with degradation monitoring; response times reduced 40%.

• Built low-level integrated systems on a Go-based monokernel — performance-critical paths under hard latency bud- gets.

Technology stack:

Platform Go, Kafka, Kubernetes, GCP, Docker

ML Forecasting, anomaly detection in production

Systems Go monokernel, gRPC, low-latency

Researcher (Formal Methods)

Université Toulouse III - Paul Sabatier

z September 2017 – February 2021 * Toulouse, France

• Designed and built a modal-logic verification platform (model checker / SAT solver) with a cross-platform UI shipped to web, iOS and Android — a 0-to-1 product from specification to app-store release.

• Published 4 papers in international conferences (JELIA, AAMAS, EMAS, PDP) on protocol verification and decidability. Technology stack:

frontend React, Elm

mobile React Native (iOS & Android releases)

backend .NET, F# Suave, Java, OCaml, CoQ

Lead Developer & Architect

Pearl traveltech

z March 2016 – June 2017 * San Francisco, USA

• Built “Permiso”, a formally-verified model checker for immigration legal requirements, deployed for 500+ business travelers — delivered as both an embeddable web component and a stand-alone system on AWS. Technology stack: Oracle; React, Elm (embeddable component); .NET, F# Suave, CoQ; AWS Technical Director & Architect

Flint Technologies

z 2007 – 2016 * London, U.K.

• Led a 52-developer organization; architect for Java/Spring BPMN platforms serving the Ministry of Economy, ISSSTE and CONEVAL — 100,000+ users — with reusable frameworks, design reviews, code reviews and mentoring as stand- ing practice.

• Delivered Magento e-commerce integrations with payment systems and ERP (London), handling 10,000+ monthly transactions.

Technology stack:

backend Java Spring, PHP Magento, .NET, SAP

db SQL Server, Oracle, Informix, Hadoop, MySQL

esb WSO2, JBPMN

Lead Architect

Wrapit PLC.

z 2005-2007 * London, U.K.

• Architect and lead developer of a C# ERP for gift-list management using an inventory-less predictive model, increas- ing operational efficiency by 50%.

CTO

Dive Consulting

z 1988-2004 * Mexico City

• Led 20+ projects across a 20-programmer team, delivering corporate solutions with a 95% success rate.

Contact this candidate