I make AI work inside enterprises that don't make it easy.

For eleven years, I've worked with brilliant engineers and business leaders to turn AI proofs of concept into systems companies actually depend on.

Carnegie Mellon '15 IBM → Gartner Denver, CO mail@lmarquina.com

35+

Production AI deployments

12M+

Subscribers on largest platform

Years in production AI

10+

Industries shipped in

Press

Build

Data architecture to ML engineering to production. The full picture, not one layer of it.

Lead

Teams of 5 to 50+. Delivery, rollout, and program execution across orgs and time zones.

Advise

Platform choices and AI governance for CIOs, backed by bake-offs and evals I run myself.

Agentic AIConversational AIVoice & NLPRAGMulti-Agent OrchestrationMLOpsAI Governance

Selected Work

View all →

Build · Telecom & Media

Tier-1 U.S. Telco

One Virtual Agent Platform, Every Channel, One Telco

IBM North America's first contact center AI program, and I ran it for eight years. My team built seven virtual agents for subscribers, field technicians and employees across chat, voice, SMS and the app, on an orchestration layer we wrote from scratch. IBM was paid per deflected interaction. We re-platformed it twice without dropping traffic. It still runs.

$350M+

Documented OPEX relief

~50%

Containment, up from ~15%

~500K

Monthly active users

2017–2025 · Telecom & Media

Build · Energy & Utilities

Agentic HR & IT Self-Service

Fortune 200 Energy Utility

HR and IT self-service for 23K employees on Now Assist, with Claude on Bedrock behind it.

40% containment · 30-point CSAT gain

2023–2025

Build · Energy & Utilities

GenAI Open Enrollment Assistant

Regional Electric Utility

IBM Consulting's first generative AI solution in production, built in early 2024.

70% containment · ~$1M/yr saved

2024

Lead · Information Services

Copilot Studio Agent Factory

Global Information Services

The controls behind a Copilot Studio rollout that grew from a few dozen agents to thousands.

800+ ungoverned agents found and remediated

2025–2026

Build · Technology

AskHR, IBM's HR Assistant

IBM

IBM's own HR assistant, rebuilt as master and child agents at .98 routing F1.

10M+ conversations a year · .98 routing F1

2020–2024

Notes

View all →

Future of Work Mar 2026

My biggest takeaway from Gartner's D&A Summit wasn't actually about AI. The expectations for the engineer of the future will shift — not just in what we build, but in how we communicate it.

AI Access Feb 2026

As foundational models get closer to our everyday life, their underlying training data creates an access schism by default. Foundational models learn a representation space whose structures are disproportionately shaped by English-language data.

Agentic AI Nov 2025

Agentic AI was the #1 topic in every CIO/CDO/architect conversation I had at Symposium. The more vendors claim to be agentic, the harder it is to find the ones that actually are.

AI Reasoning Oct 2025

OpenAI researchers claimed GPT-5 had "solved" a batch of Erdos problems. Turns out… not quite. Retrieval ≠ Discovery. Finding an answer isn't the same as creating one.

Writing

View all →

EmployeeWorks & Otto: ServiceNow's Foray into the Employee AI Front Door

Otto and EmployeeWorks give ServiceNow a real conversational AI stack. A practitioner's take on the good, the gaps, and what to ask before you migrate.

lmarquina.com

ServiceNow Generative AI: Now Assist

A practical breakdown of Now Assist and what it means for enterprise AI delivery in service workflows.

Medium

Path

2026

Guest lecturer, AI & Innovation · Carnegie Mellon

2025 –

Partner & AI Practice Leader · Gartner

2015–25

Data Scientist → Assoc. Partner · IBM

2016

Olli — Autonomous Shuttle Voice Assistant (IBM)

2015

B.S. Statistics, Honors · Carnegie Mellon

Let's talk.

Working on AI that has to hold up in production? I'm always up for comparing notes.

Email me LinkedIn