Karthikeyan Selvam — AI Product & Enablement Leader

Act I — Origin · 2007

Every AI story I trust starts in production. Mine starts on an AS/400.

Tata Consultancy Services. Sales-audit reconciliation posting to Oracle — millions moving nightly, and the books had to balance by morning. No demos, no slideware. Just systems that either worked or didn't. That standard never left.

tcs — session · 2007

> login: karthikeyan.s · chennai

> platform: IBM AS/400 · sales-audit recon

> job: nightly batch → oracle · zero-error tolerance

> status: balanced by morning. every morning.

> lesson_01: trust is earned in production

Act II — The Climb · 2011–2023

Sixteen years of shipping things that had to work.

Not a straight line — a widening one. Data, then automation, then programs measured in tens of thousands of devices. Follow the ember.

2011 · HEWLETT-PACKARD

Learning to read an enterprise through its data

Assistant Manager — data analysis, process assessment, functional design, change controls. Where process discipline was built.

data & process

2015 · ATOS — RPA LEAD

Teaching software to do the work

50+ RPA automations across HR, Finance, and Procurement — leading consultants, business analysts, and developers.

50+ automations

NATIONAL GRID — AIOPS

Predictive operations across a 40,000-device estate

AIOps (Nexthink / ITOM) plus enterprise backup & recovery — turning reactive IT operations into measured, proactive ones.

40,000-device estate

MICROSOFT — COPILOT FOR SECURITY

Grading an AI before the world used it

My first direct step into AI — straight off the AIOps program. Rubric-driven evaluation of Microsoft Copilot (Entra) with Microsoft Engineering: accuracy, correctness, user satisfaction. The score moved 20%.

+20% RSAT · first AI role

ATOS — PROGRAM DELIVERY · THE INTERIM

12,000 devices, moved with AI in the toolkit

In the interim, a Windows 11 migration owned end to end — with AI-assisted analytics sharpening readiness assessment and adoption strategy. Delivery at scale, now AI-informed.

12,000 devices

Then the industry turned.

Enterprises stopped asking whether AI works…

…and started asking who can make it work in production.

So I built the answer.
— and I run it every day.

Act III — The Turn · Live from production

Not a case study. A system I designed, govern, and run today.

An enterprise multi-agent AI platform on Google Cloud, powered by Vertex AI / Gemini — delivering real consulting workflows with governance and cost discipline engineered in.

Platform architecture — running now google cloud · serverless · regional-only

governance built in: auth-gated · 0 exposed endpoints · 100% audit-traceable outputs · scale-to-zero · automated model-lifecycle steward · 1-command idempotent deploy

Meanwhile — what I lead now

Three disciplines, one throughline.

Getting AI and automation adopted, governed, and paying for itself.

AI Product & Platforms

Founding, building, and running AI products — roadmap to release. Multi-agent systems, LLM evaluation programs, governed compliance agents, and the deploy pipelines behind them.

Where: Multi-agent AI platform · Copilot for Security · RevokAI

AI Enablement & Adoption

Moving organizations from AI curiosity to dependable, everyday use — structured enablement, peer champions, leaders modeling the change, and adoption that's measured, not assumed.

Where: Copilot Enablement Playbook · AI Mission Challenge 2025 winner

Enterprise Program Delivery

Programs from $200K to $20M, teams of up to 20 — migrations, AIOps, backup & recovery, and RPA at scale, owned end to end from budget to executive stakeholder.

Where: Windows 11 (12,000 devices) · National Grid AIOps (40,000) · 50+ RPA automations

Act IV — Shipped

Proof, not promises.

AI systems I've founded and run — from a production multi-agent platform on GCP to a governed compliance control on the Microsoft stack. Each card opens an interactive story with the decisions behind it.

keep scrolling — the shelf slides→

01 / 06Production

Multi-Agent AI Platform

An enterprise AI platform delivering consulting workflows — specialized agents on a serverless GCP stack, hardened against strict org policies, with governance and cost engineering built in.

FastAPI · Cloud Run · Vertex AI · FirestoreView story →

02 / 06Production

Change Management Agent

Generates a complete OCM deliverable set — 4 phases, 18 deliverables — from a formal methodology, with version-tagged, audit-traceable outputs and branded PDF export.

Python · Vertex AI · Versioned skill engineView story →

03 / 06Production

License Right-Sizing Agent

Reads real Google Workspace usage across a tenant and surfaces defensible savings — with a zero-usage guard so quiet-but-active users are never wrongly downgraded.

Admin SDK · Reports API · Cloud Run JobsView story →

04 / 06Production

Infra Health Check Agent

Automated infrastructure health monitoring that produces shippable PDF reports and emails them out — replacing manual review cycles with a repeatable pipeline.

Python · Analysis pipeline · Email automationView story →

05 / 06Pilot

RevokAI — Governed Access Control

An AI compliance control for access revocation: AI correlates the hard identity matches a report can't, deterministic rules plus mandatory human approval decide, and every action lands in an exportable audit trail.

Copilot Studio · Power Automate · Office ScriptsView story →

06 / 06Production

Customer Outreach Agent

Autonomously researches a customer's strategic priorities with live web search, scores alignment against a service portfolio, and drafts personalized outreach — only opportunities that clear a strict evidence threshold are ever actioned.

Vertex AI (Gemini) · Live research · Scoring engineView story →

Act V — How I operate

The operating system behind every engagement.

Pilot to production without the stall.

/01

Evaluate before you ship

Rubric-driven LLM evaluation — accuracy, technical correctness, user satisfaction — before anything reaches users. It's how a Microsoft product metric moved 20%.

/02

Governance is a feature

Role-based access, audit-traceable outputs, and an automated model-lifecycle steward that migrates agents off retired LLMs — compliance without manual toil.

/03

Adoption is the product

Technology is 30% of a transformation. Readiness measurement, communications, and executive alignment are what make the other 70% land.

/04

Cost is architecture

Scale-to-zero services, token budgeting per output, regional-only infrastructure. Efficiency is engineered in from day one — the same discipline whether the program is $200K or $20M.

The Copilot Enablement Playbook

My structured 12-week program for moving an organization from Copilot curiosity to dependable, everyday use — because access is no longer the constraint; intentional adoption is.

01 · BUILD

Practical confidence

Copilot introduced through the work people already do — email, meetings, documents — including where it fails, how to guide it without over-engineering prompts, and how enterprise data shapes results.

02 · ACTION

Real work, live

Copilot applied against live emails, documents, and deliverables in small-group sessions that surface real friction points, encourage shared learning, and normalize AI-assisted work within teams.

03 · SUSTAIN

Make it stick

Credible peer champions, leaders modeling usage visibly, reinforcement through relevant examples — adoption sustained by influence rather than enforcement, measured and adjusted as it grows.

12-WEEK ROADMAP · Mobilization (wk 1–2) → Skill Foundation (3–6) → Embedded Enablement (6–9) → Review & Direction (10–12)

Recognition — earned in the field

Winner — Atos AI Mission Challenge 2025

"MOST IMPACTFUL BUSINESS SOLUTION" · SOLVED WITH MICROSOFT COPILOT

Bronze — The Copilot Games 2024

TEAM DWP NORTH AMERICA · ATOS

Epilogue — your move

You've read the story. Now interrogate it.

Consulting engagement, a leadership conversation, or a hard question about where enterprise AI is really headed — I'm open to all three.

Open to: AI Program Management · AI Adoption · AI Enablement · AI Product Leadership

Connect on LinkedIn skarthik1710@gmail.com

The ember that's been following you down the page? It's in safe flippers — Skipper, my AI, is holding it at the bottom-right, one tap away on every page. Runs on my own Firebase + Gemini stack. The site is the demo.

I take enterprise AI from slideware to shipped.