GoldenArc
Fractional AI leadership · San Francisco Bay Area

Your AI demo impressed the board.
Production is where it’s dying.

I take one stuck or flaky AI initiative from demo to dependable production in 90 days — shipped, instrumented, and cost-controlled. I’m Mayur Sethi: an operator who builds and runs production AI systems with my own P&L on the line. You get working Human+AI systems. Not a deck.

✓ Fixed fees, scoped in one call✓ One real fix shipped inside the first 3 weeks✓ You own every line of code and every eval
20 years building & delivering at
HPFujitsuHCLCognizantHitachiVeritasDataStax
The gap nobody owns

95% of AI initiatives never make it past pilot. The reasons are boringly consistent.

MIT found only ~5% of enterprise AI pilots produce real financial gains. Gartner expects 40%+ of agentic AI projects to be cancelled by 2027 — on cost and reliability, not ambition. The gap between an impressive demo and a dependable product is an engineering and operating discipline. Most teams have neither the time nor the scar tissue to build it.

Reliability nobody owns

The demo works in the meeting and fails on real load. Edge cases, hallucinations, and silent regressions land on whoever shouted last.

Costs rising without a ceiling

Inference spend creeps toward 20%+ of revenue for AI-first products. No per-call cost visibility, no budget owner, no kill switch.

No evals, no quality bar

“It seems better” is not a release criterion. Without an eval harness and human-in-the-loop review, every change is a gamble.

Leadership flying blind

Is this a 3-week fix or a 3-quarter rebuild? When nobody can answer, the initiative stalls — and the team quietly loses the room.

How engagements work

Two steps. Both fixed fee. Both ship working software.

No six-month discovery. No strategy deck you could have written yourselves. The audit de-risks the build; the build ships the outcome.

Step 1 · 3 weeks · Fixed fee

AI Production Readiness Audit

A full map of one stuck AI feature — where it breaks, what it costs per call, where it won’t scale. Not a paper exercise: I ship a real fix inside the audit.

  • — Failure-path map of one AI feature, end to end
  • One shipped win: the worst failure path made dependable, inside 3 weeks
  • — A plain-language definition of “production-ready” for your feature
  • — Eval + quality bar your team keeps forever
  • — Prioritized 90-day build roadmap: reliability, cost, scale, data plumbing

Fee fully credited toward the Launchpad. And if it’s truly a 3-quarter rebuild, you’ll know in week one — and you can stop there.

What happens after

Most clients don’t want the discipline to leave when I do. Many keep me on as their Fractional Head of AI — ongoing ownership of evals, cost control, and build/buy/kill decisions across their AI surface. That’s a conversation we have during the work, once you’ve seen how I operate — not a pricing page.

Who this is for

I do my best work for a narrow kind of company.

A strong fit if you are…

  • — A funded or profitable B2B company — typically $10M+ ARR or Series B and beyond
  • — Already shipped (or attempted) AI features — and now stuck on reliability, cost, or scale
  • — Knowledge-work heavy: B2B SaaS, agencies, services firms with real workflows to automate
  • — Ready to give an operator real access — code, data, and an hour a week of decision-maker time

Not a fit if you need…

  • — A dev shop to build your first product from scratch — I harden and ship what exists
  • — A strategy deck for the board — I ship systems; the deck writes itself afterward
  • — Help before you have revenue or users — pre-revenue teams have cheaper options
  • — Someone to own AI forever without your team learning it — I hand over, always
The 90-day journey

Day 1 starts where the audit ended. Running.

By the time the Launchpad begins, the failure map is drawn, access is live, the first fix has already shipped, and the roadmap is prioritized. There is no discovery phase — there is only the build. Your team sees progress weekly, not at the end.

Day 1
Build begins — roadmap in hand

First sprint scoped to the highest-ROI failure path. No discovery phase.

Day 7
First hardened path in staging

Measured against the eval bar set in the audit.

Day 14
First production deploy

Behind a flag. Real traffic, controlled blast radius.

Day 30
Eval harness gating releases

Quality scores decide what ships — not vibes.

Day 60
Cost ceilings and scale, proven

Per-call economics visible. Load proven, observability live.

Day 90
Dependable production, handed over

Instrumented, cost-controlled, yours.

Start with the audit →The journey begins where the audit ends.
Why an operator, not an agency

Agencies bill hours. Consultants write decks. Operators are accountable for what runs in production.

I’m not a dev shop with an AI page, and I’m not a strategy firm that hands you a roadmap and leaves before the hard part. I run AI-first businesses with my own P&L — the same systems, evals, and cost discipline I sell are the ones keeping my companies alive.

−20% → +15%
EBITDA swing in 12 months

As CRO of a healthcare services company, I rebuilt operations AI-first — 8+ FTE functions now run by 2 people + 12 AI agents. Opex down 40%.

10 days
Whiteboard → MVP

EcomDataIQ, my own AI analytics platform: live on GCP with paying customers, $10M+ of client revenue under management.

$1.6B
Acquisition positioned

At DataStax I built the IBM partnership from zero to ~30% of ARR — the strategic thesis behind the ~$1.6B IBM acquisition.

19% EBITDA
Bootstrapped, profitably

I co-founded Life Sutra: multi-million revenue in two years, selling through Amazon, Walmart, and Macy's.

How I work — the operating principles

  • Evals before features. If you can't measure quality, you can't ship changes. The harness comes first.
  • Cost is a product requirement. Every AI call has a price; every feature gets a ceiling and an owner.
  • Humans in the loop, by design. The goal is dependable Human+AI systems — not unsupervised magic.
  • Narrow first, then wide. One initiative, shipped and measured, beats five pilots in flight.
  • Your team owns everything. Code, evals, dashboards, decisions. I make myself unnecessary.

Agency vs. consultancy vs. GoldenArc

Dev agencyStrategy firmGoldenArc
DeliverableBillable hoursA deckWorking production systems
IncentiveMore hoursMore scopeFixed fee, shipped outcome
First proofSprint 6+Never runsA real fix in week 3
Runs AI in production with own P&LRarelyNoYes — currently, daily
Mayur’s photo
San Francisco Bay Area
Stanford Executive AI Product Management
Google Generative AI Leader
The principal

Mayur Sethi

Founder & Principal, GoldenArc · Fractional Head of AI

Twenty years across enterprise tech — and one consistent pattern: it ships. I built 0→1 products at HP and HCL (the SAP-as-a-Service blueprints I co-designed now generate $1B+/year for HCL). I delivered at enterprise scale at Cognizant, growing a single engagement from $90M to $280M ARR for clients like JP Morgan and BlackRock. I sold to the Fortune 500 at Hitachi and Veritas — ~$60M from 17 new accounts, 200% YoY partner growth. And I positioned DataStax for its ~$1.6B IBM acquisition by building the IBM alliance from zero to ~30% of ARR.

Then I came back to building. Today I run production AI with my own money on the line: a live SaaS platform on GCP with paying customers, multi-agent systems, custom models, and human-in-the-loop evaluation frameworks. I close the gap most AI teams can’t — taking AI from impressive demo to dependable, cost-controlled production.

HP · Solution ArchitectHCL · Sr. Product ManagerCognizant · $90M→$280MHitachi · Director GSIVeritas · DirectorDataStax · Head of Alliances
Straight answers

The questions CTOs actually ask

Who actually does the work?+

Me — hands-on, in your codebase, alongside your team. No bait-and-switch to a junior bench. Where specialist depth helps, a small vetted network plugs in under my direction.

What do you need from us?+

Repo and infra access, a data sample, and one hour a week with a decision-maker. If you can't give an operator access, you're not ready for an operator.

What if it really is a 3-quarter rebuild?+

Then you'll hear that in week one, in plain language, with the evidence — and you can stop at the audit. You keep the map, the eval bar, and the shipped fix either way.

Which stacks and models do you work with?+

Model-agnostic by principle: OpenAI, Anthropic, Gemini, open-weights — whatever the evals and unit economics justify. Deep hands-on with GCP (Cloud Run, BigQuery), comfortable across AWS and Azure.

One call, real answers

You'll leave the first call knowing whether this is a 3-week fix or a 3-quarter rebuild.

Thirty minutes, no pitch theater. Bring your stuck initiative; leave with an honest read on what it takes. The audit is a fixed fee scoped on that call — a fraction of one senior AI hire, with the first fix shipped in three weeks.

Every note is read by me, not a funnel. — Mayur Sethi