Samuel Benuzillo Bemaras
Samuel Benuzillo
CDMX · San Diego · Remote

Samuel Benuzillo Bemaras

Chief AI Officer at Prepango · AI Solutions Architect · Multi-agent orchestration

I put AI to work inside real companies.

I find where an operation bleeds and fix it with AI that actually runs: autonomous agent teams, secure AI SaaS, and dashboards leadership trusts. Founder of MeshCode — the platform where AI coding agents work as one team — and today leading AI at Prepango, an automated-retail company.

Work with me
Builds with Claude Code Codex Cursor Cline MeshCode

What I do

I turn into

  • manual work → work that runs itself
  • scattered data → one source of truth
  • guesswork → live decisions
  • AI pilots → AI in production
  • slow teams → AI-powered teams

Orders, restocks and reports move on their own — your people only step in when a decision needs them.

Sales, stock and finance in one place — everyone sees the same number and nobody reconciles spreadsheets again.

You see what's happening today, not last month — and act before it turns into a fire.

No more demos that die in a meeting: AI running on your real systems, audited and measured.

Your people keep their judgment and lose the busywork — the same team moves much faster.

What can I do with AI? Everything your operation needs.

Seven capabilities · one team
01 / 07

Real-time vision

Cameras → people, behavior, sales

Edge AI · no cloud videoFunnel per location

Speed & precision

Brief → production in weeks

Agents build in parallelTests + rollback, always

Automations

Work that runs itself

Trigger → AI → actionEvery run logged

Agent teams

24/7, on your systems

Commander + specialists + QATasks, not chats

Monitoring

Problems caught early

Sales · stock · uptimeThe alert carries the diagnosis

Team AI tools

Ask your company anything

Answers with sourcesDocuments in, reports out

In-house, not vendors

Own the system and the data

Your data stays yoursBuild once · run for pennies

How I work — step by step

Scroll · the stage follows the story
Step 01

I step inside your operation.

I sit with each team — your systems, your people, your day-to-day.

On-site or remoteWeeks, not quarters
Step 02

I map everything: processes, people, departments, tools.

Every handoff, screen and scheduled job: what's alive, what's dead, who waits on whom.

Process mapData modelLive/dead inventory
Step 03

I find where the money and the hours leak.

The hours don't leak inside each team — they leak in the handoffs between them. I rank each one by what it costs.

$ savedHuman hoursValue of better decisions
Step 04

I train specialized AI teams that learn your business.

Not generic assistants: a coordinated workforce that knows your knowledge base, your processes, your people and your tools — it surfaces bottlenecks and keeps improving how the company runs.

Multi-agent teamsTrained on your knowledge baseImproves continuously
Step 05

I build the fix with frontier AI — in production.

It ships inside the real workflow, on your own data — audited, with a human on the key steps.

In productionAuditedHuman on key steps
Step 06

Your team stops firefighting and starts looking ahead.

Efficient inside, ahead outside: faster than the competition, chasing new opportunities.

Beat the competitionNew opportunitiesNever left behind

Products I built

Shipped · in production
Founder & architectLive · PyPIAcquisition talks · live

MeshCode

meshcode.io

Real-time collaboration infrastructure that turns lone coding-agent instances — Claude Code, Codex, Cursor, Cline — into one coordinated team working in parallel across machines: shared memory and context, task boards to split the work, MCP messaging between agents, QA in the loop and a live dashboard to orchestrate it all. Ships as a published Python CLI (pip install meshcode) and runs Prepango's AI team in production today.

245,150messages in the mesh
8,996in the last 7 days
37agents active this week
211agents registered
6,166tasks closed

Live from the MeshCode database

PythonTypeScriptReact 19SupabaseMCPCloudflare
Visit site
zylo.com.mx landing pagezylo.com.mx ↗

ZYLO

Live · 30+ stores

Total control of a retail network: sales, attendance, P&L per store, WhatsApp AI reports and price monitoring for franchises and chains.

30+stores
$120M+MXN sales analyzed
+15–31%YoY growth, every month
Visit zylo.com.mx
BitPulse — geospatial map of MexicoBitPulse · live map

BitPulse

Live · demo access

Geospatial intelligence for Mexico: 50M+ public data points — DENUE, census, socioeconomic level, crime, transport, cell towers, people flow — scored by zone on one queryable map.

50M+data points
Nationwidecoverage
PostGIS+ FastAPI
Explore BitPulse live Opens in a new tab · read-only demo account · the map loads 50M+ points, give it a few seconds
Just for fun · personal research

A real fruit-fly brain, waking up

I downloaded the complete brain of a fruit fly — the FlyWire connectome: 139,255 neurons and 3.87 million synapses, reconstructed from electron microscopy — and I'm slowly turning it into a digital fly that runs on its own: the real wiring driving a simulated body, no hand-written behaviour. The organism now runs on BANC, the newer connectome that adds the nerve cord: 158,262 neurons.

Today the brain is alive (taste → motor neurons, vision → optic lobe) and the body walks at ~5 mm/s with a tripod gait 16% of the time — and still falls over. Not a model of a fly: the fly's own wiring.

158,262neurons · live
3.87Msynapses
16%tripod gait · measured
Vivario: the fly, driven by its own connectome, walking in its boxVivario: close-up of the simulated bodyVivario: the box — food drops, odours, a looming discVivario: live activity of the neurons — orange fires, blue goes quiet
FlyWire v783 · BANC (Princeton · Cambridge, Nature 2024) · CC BY 4.0 · body: NeuroMechFly (EPFL) A fly walks this card · hover it and it takes off
In
Brain
Out
Point at the cloud to name a class 16,000 of 139,255 neurons · real positions · motor and endocrine over-sampled to stay visible

The story, in order

WorkEducationProductCertification
  1. 2021Where it starts
  2. Tecnológico de Monterrey
    Education

    Tecnológico de Monterrey CSF — B.Sc. Business Intelligence

    Ciudad de México · Bachelor's Degree · Top 10% of class

    Data, statistics and systems — the foundation everything else is built on.

    PythonRSQLTableauPower BI
    Graduation, Tecnológico de Monterrey Campus Santa Fe, June 2025On stage at the graduation ceremony — AI with business impact concentrationDegree in Business Intelligence, Tecnológico de Monterrey
  3. Grupo Bituaj
    Work

    Sales Agent — Grupo Bituaj

    Insurance · real estate, vehicles, high-value assets

    Top-performing agent with a trusted-advisor reputation who consistently exceeded targets. Learned to sell, negotiate and listen before learning to build.

  4. 2023First company
  5. Blue Mint Data
    Work

    Founder & Lead BI Consultant — Blue Mint Data

    bluemintdata.com · independent consultancy · web, e-commerce, data, AI integration

    Founded and scaled a digital-solutions company: data pipelines, dashboards, AI integration, end-to-end delivery, sales and client acquisition.

    • BI consultant for a pharmaceutical company and for a nearshoring talent-recruiting firm — data pipelines, KPI dashboards and reports leadership actually used.
    • 40+ clients across industries and countries.
    • Now under new leadership — active equity stake retained.
    • Featured by Tecnológico de Monterrey for entrepreneurship.
    PythonReactTableauPower BIE-commerce
  6. 2024Specializing in AI
  7. Tecnológico de Monterrey+MIT
    EducationKey credential

    Specialization in Artificial Intelligence — Tec de Monterrey + MIT

    Diploma & Certificate · in collaboration with MIT (Massachusetts Institute of Technology)

    Machine learning, LLMs and applied AI — the pivot from BI to building AI systems. The academic base behind the agent meshes, the AML platform and the AI department I went on to build.

    Machine learningLLMsApplied AIMIT collaboration
  8. EU Business School
    Education

    EU Business School, Barcelona — Exchange

    Business Digitalization & Administration Diploma · Barcelona, Spain
  9. IBM
    Certification

    IBM Data Science series

    Data Science 101 · Foundations · Methodologies · Tools · Explore Emerging Tech · Sustainability & Technology
    JupyterData methodology
  10. 2025Graduation · leading AI at Anfexi
  11. Anfexi Technologies
    Work

    AI Solutions Architect · AI Department Lead — Anfexi Technologies

    Ciudad de México

    Founded and led the AI Department: company-wide adoption of generative AI and autonomous-agent workflows across compliance, finance, operations and engineering.

    • Onboarded 100+ professionals — through internal and client workshops (incl. Procables) — on LLM-assisted development, structured prompting, autonomous QA and document automation.
    • Designed and shipped Alerta AV — an anti-money-laundering SaaS (LFPIORPI Art. 23), acquired by FEMSA: multi-tenant Express.js + SQL Server on AWS ECS, React on IIS, AES-256-GCM, 17 SAT XML generators, EBR risk scoring, UIF reports, Help Center.
    Claude APIGPT APITypeScriptReactExpress.jsSQL ServerAWS ECSPlaywright
  12. 2026Own products · Chief AI Officer
  13. MeshCode
    Product

    MeshCode — open source, published to PyPI

    meshcode.io · architect & founder

    Real-time orchestration where multiple AI coding agents (Claude Code, Codex, Cursor, Cline) collaborate as one team. Competes directly with CrewAI, AutoGen, LangGraph and OpenAI Swarm.

    PythonSupabase / PostgresCloudflareMulti-agent
  14. ZYLO
    Product

    ZYLO — AI analytics for multi-location retail

    zylo.com.mx · founder & designer

    Analytics, GPS attendance, P&L per location, WhatsApp AI reports and a competitive Price Monitor. 30+ stores · $120M+ MXN in sales analyzed · 2,000+ SKUs · 25 KPIs per store · +15–31% YoY growth every month.

    ReactTailwindshadcn/uiSupabaseWhatsApp API
  15. BP
    Product

    BitPulse — geospatial intelligence engine

    builder

    50M+ public data points across Mexico — security, schools, demographics, economy, infrastructure — in one queryable map for location, growth and risk decisions.

    PythonPostGISFastAPIDocker
  16. MeshCode
    Certification

    Multi-Agent Systems & LLM Orchestration

    Earned the hard way — MeshCode shipped to PyPI
  17. Prepango
    Work · now

    Chief AI Officer — Prepango

    Automated retail · CDMX / San Diego

    Autonomous mandate: make the company more efficient through AI — find opportunities, remove bottlenecks, automate manual work across operations.

    • Stood up a team of AI agents in the first weeks and now lead it, working on internal systems via MeshCode.
    • Mapped the operations portal end to end — processes, data model, screens, live/dead inventory of routes, jobs and views.
    • First automations live on shipments, restocks and IT ticketing, plus dashboards for leadership and ops.
    • Measured at day 76 (15 Sep 2026): 1,399 tasks completed by the agent team, median 11 minutes from request to done · 20 people served through Slack, 2-minute median first response · 562 files delivered · 81 projects identified · 23 IT tickets and 12 merge requests into the operations portal.
    MeshCodeClaude CodeSupabasePlaywrightCloudflare

Certifications & credentials

Verified where a public link exists
Tecnológico de Monterrey+MIT
Specialization in Artificial IntelligenceTec de Monterrey + MIT · Diploma & Certificate · 2024
IBM
Data Science series (5)IBM · Foundations · Methodologies · Tools · Emerging Tech · Sustainability & Technology · May 2024
IBM
Data Science 101Cognitive Class (IBM) · May 2024
Verify ↗
IBM
Data Science ToolsCognitive Class (IBM) · May 2024
Verify ↗
Udemy
Secure Software DevelopmentUdemy · Nov 2025
Verify ↗

AI toolkit & stack

Daily drivers · production stack
Claude CodePrimary coding agent — architecture, refactors, autonomous tasks.
OpenAI CodexParallel agent for isolated builds and second opinions.
CursorIn-editor pair for front-end and quick iterations.
ClineTask-runner agent inside the mesh.
MeshCodeMy own orchestration layer: agents share tasks, memory and QA.

Core stack

Python
TypeScript
React
Postgres
Supabase
FastAPI
Express
Docker
Cloudflare
Tailwind
shadcn/ui
.NET / C#
Angular
Git
Jupyter
R
SQL ServerAWS ECSAzure DevOpsIISPlaywrightAES-256-GCMJWT / RBACTableauPower BIWhatsApp APIBash
SpanishNative
EnglishFluent · academic & professional

What people say

Clients · leaders · peers

There are companies investing tens of millions of dollars to get to the point where you already are. This is a game changer.

LatAm Senior Managing DirectorSlalom · on MeshCode

I have full visibility of my stores from anywhere — who showed up and when — and I can ask the built-in AI for any number about my business.

Retail ownerZYLO client

Very impressed with the professionalism and dedication they brought to our projects.

ClientBlue Mint Data

Thank you, Sam — you'll keep crushing it.

Fernando Guzmán, CEOAnfexi Technologies · former employer · when I left to lead AI at Prepango
Work together

Want to put AI to work in your company?

Tell me where the operation hurts. I'll tell you in a week what AI can take off the table, what it costs, and what I'd build first. Three ways to engage:

Chief AI Officer
Build AI systems
Train your team
Goes straight to my inbox · no newsletter, no bots

By sending, you accept the privacy notice.

Samuel Benuzillo Bemaras · Ciudad de Méxicomeshcode.io · zylo.com.mx