Marcelo García — AI Founding Engineer

// Track record — real companies, real numbers

Six years turning attention into revenue.

I've owned CRM & retention for regulated sportsbook and casino operators across Latin America — the commercial side that actually moves GGR. Here's the receipts.

Platforms, operators & brands I've worked with and for

Head of CRM · 2025–2026

I led retention for a real conversational sportsbook.

ChatBet lets bettors place bets inside WhatsApp & Telegram — natural language turned into a betslip on the operator's sportsbook. I owned player lifecycle, segmentation and retention strategy across live LatAm operators, on a platform integrated with Kambi and Digitain and deployed with BetPlay, betvip and Betcris.

WhatsApp / Telegram Kambi + Digitain integrated LatAm operators live

chatbet.gg ↗

ChatBet Proven Success: +20% monthly GGR, +30% 90-day retention, +35% new bettors, +30% average wager size

ChatBet's own published operator results — chatbet.gg

Founder · CRM Director

ChatPuesta

2026 – Present · current venture

My own conversational CRM for LatAm sportsbook players: segmented, fixture-triggered campaigns over WhatsApp/Telegram, with in-chat bet placement. Early-stage — building the retention engine and the product at the same time.

Conversational agentWhatsApp API

Head of CRM

ChatBet

2025 – 2026

Led CRM, engagement & retention for the conversational betting platform above — segmentation, reactivation and bonus mechanics across operators running both sportsbook and casino, with a WhatsApp Business (Meta) delivery layer.

+30% retention*+20% GGR*Kambi · Digitain

CTO → Senior CRM

Unifymedia · Playgreen

2023 – 2025

Shipped a WhatsApp betting product with the operator's CPO and ran mass WhatsApp campaigns timed to kickoff and in-play moments of live football — a repeatable weekly format that pulled players in at peak intent.

Product + CRMEvent-triggered

CMO · CRM Director

ACTA Managers

2023 – Present

Drove managed creators past six figures in cumulative revenue with AI-driven content — and built ACTA OS, the full internal ERP the agency runs on (see below). Growth and engineering, same person.

Full-stack ERPAI content ops

Creative + Strategy

MelBet & 1XBET · via BCKBN Social

Campaigns

Produced influencer video content and the full implementation plan for MelBet's UGC programme, and designed a 3-tier influencer acquisition campaign for 1XBET — two of the biggest betting brands in the world.

UGC programmeInfluencer acquisition

The through-line

I think like the funnel

I register at operators as a real player, go silent for six days, screenshot every lifecycle message they send — and don't — then hand the Head of CRM a one-page audit: "your first 72 hours are leaking players." Retention is a reflex.

Retention obsessed

* +30% 90-day retention and +20% monthly GGR are ChatBet's own published operator results (the "Proven Success" figures on chatbet.gg), attributed to the platform — not claimed as my personal numbers. I never invent a metric.

// The AI Founding Engineer questions, answered with evidence

Your application form is my job description.

The retention track record is one half of me. The other half: I read your form's questions closely, and I've already shipped working systems against each one.

◆

"An agent with dozens of tools acting on real accounts & services"

JobHunt OS acts on my Gmail (IMAP), 13 different ATS platforms, and a live browser session. ZEBRA acts on Instagram, YouTube and the Meta API. I've been building exactly this class of agent for months.

◆

"The hard part is deciding what to say when nobody asked"

My Command Center is a proactive agent: it decides on its own when a failure, a deadline, or a shipped result is worth a Telegram ping — and when silence is the right call. Signal over noise, by design.

◆

"Where do you draw the line on agents acting without approval?"

I've drawn it in production. JobHunt OS never fabricates a screening answer, never bypasses a CAPTCHA, caps applications per company, and routes anything irreversible to me for a one-tap approval on my phone. Autonomy with brakes.

◆

"Demo works, production fails — LLM failure modes"

I treat verification as sacred: an application is only "confirmed" when a real inbox email proves it. Evidence for every action (2,500+ screenshots). I've debugged the regex-OOM, the empty-SPA render, the silently-miscoring bug — from production data.

◆

"You ask Claude for a screen and it's technically correct but generic"

I tell it to stop designing like a template: commit to one strong reference, kill the defaults, name the exact type ramp, spacing scale and one signature interaction — then iterate on the seams. This very site is the artifact of that loop.

◆

"What are you building right now, outside of work?"

A dozen live systems — an AI video factory (reelforge), a Shopify competitive-intel crawler (shopscope), a multi-account lead listener (telelistener), and this portfolio, built by a 7-agent research harness that mined 3 years of my own work.

// The one that proves the range

I ran a company with an AI executive team.

ACTA OS is the full-stack operations platform I built for a creator-management agency (~$30K/mo, signed creators, a distributed human team). Clearest proof that I can take a real business and build the entire agentic system it runs on — alone.

Prisma / PostgreSQL data models

Commits on the staging repo in ~6 weeks

AI executives: CEO, CMO & CTO agents

Signed Electron desktop-app builds shipped

The AI executive team

I ran the org through Paperclip with CEO, CMO and CTO agents. The CTO agent shipped code through feature-branch PRs against a protected staging repo that I visually reviewed before merging to production. Agents that build, guardrailed by a human.

Three portals, 232 backend files

An Admin Command Center (40+ pages), an Employee Workspace (15 role-specific dashboards), and a Model Portal (earnings, invoices, content, tickets). Express + React 19 + Vite + Tailwind 4, on Railway, Neon & Render.

Real-workforce automation

A signed Electron app drives anti-fingerprint browsers so staff operate managed accounts without ever seeing passwords; a 17-day account-warming state machine; clock-in gated on live screen-share presence with idle detection.

A dozen integrations, wired solo

Discord.js, Telegram Bot API, Google Drive + Calendar, Calendly, Recurrente payments, Gemini for morning briefs & AI captions, Bull/Redis queues. This is what "dozens of tools acting on real services" looks like when I build it.

// Why me, specifically

A builder who actually gets growth.

Short and specific — the way you asked for it.

✓

Zero-to-one is my normal

I've stood up CRM functions from first principles and built the software they run on, with no team and no budget. That's the founding-engineer job, and I've done it repeatedly.

✓

Grind + taste + Claude Code

Exactly the stack you named. Agentic coding isn't a tool I tried — it's how I ship every day, across a dozen live systems. Speed is my edge; JobHunt OS was idea-to-verified in a week.

✓

Product + growth are one instinct

Cal AI won because they weren't separate. I live at that seam: I understand the user and the funnel, and I write the code. Beyond the chatbox is exactly where I want to be.

✓

I ship things that run without me

Self-healing services, idempotent pipelines, real verification, evidence for every action. I build for production and for the on-call at 3am, not for the demo.

✓

Credentials-out, output-in

I did two years of a CS degree and left to build. Like you, I bet on shipped work over pedigree. This whole page is the argument, made in the medium.

✓

I'll relocate and go all-in

Guatemala City → anywhere your small team is. Well-funded, stealth, shipping to real users, hardware on the way — that's the room I want to be in from day one.

// Your last question

"Why work on an AI product you can't see yet?"

Because I've spent three years quietly building the exact thing you're now funding — agents that act on real accounts — with nobody watching and no equity in the upside. You're offering the one thing I've been missing: the best team in the world at this, a real product, and a reason to go all the way. I don't need to see it. I already believe in the shape of it.

marceloandre.garcia@gmail.com LinkedIn GitHub

P.S. — I built this site the way I'd build for you: me, Claude Code, and a 7-agent research harness that mined three years of my own projects, verified every company claim on the open web, and cut the assets. The intro you (maybe) skipped was real. The portfolio is the proof of work.

I build agents that act on real accounts.