Creative Director · Florianópolis · Latam

I direct AI
like a creative team.
Not like a tool.

I'm Fernando. I lead the Latam creative team at Valtech, and I build AI agents that ship on-brand creative work at the speed real teams need. The work below is real — running in production for brands like Sebrae, Sanofi and CBMM.

Shipping work for

Selected work

Five cases that show how I direct AI through real creative problems.

01 Sebrae 2026

An image agent in the Sebrae standard.

Sebrae's creative team was losing hours every week to stock-image curation that still didn't match the brandbook. I built a Claude project — "Sebrae Image Prompt" — that takes a one-line request and produces a photorealistic image faithful to the brand and to Brazilian reality.

  • 100%fidelity to brandbook
  • 0hstock curation time
  • 1simple prompt → final image
Input from the team

"mecânico trabalhando em uma oficina de bairro"

Generated by the agent

Photorealistic documentary-style medium shot of a mechanic repairing an older compact hatchback in a small Brazilian neighborhood auto repair shop, open roll-up garage door facing a sunny street, masonry walls, worn concrete floor with oil marks, metal shelves, used tires…

Stack: Claude Projects · Nano Banana 2 (image engine) · brandbook ingestion · prompt-engineering for photographic direction.

Florist serving a customer in a Brazilian neighborhood flower shop
Popcorn vendor in a Brazilian town square with children and a dog Confectioner rolling brigadeiros in a cozy Brazilian sweet shop Mechanic repairing a car in a small Brazilian neighborhood auto repair shop
02 Sebrae 2026

A second agent — this time for illustration.

Same problem, different craft. Without a dedicated illustrator, the team relied on stock illustrations that never quite landed. I built a parallel Claude project — "Sebrae Illustration Prompt" — that decodes the brand's linework, palette and composition into prompts a generative model can execute.

  • 4image engines tested
  • GPT Image 2selected for stylistic control
  • 80+on-brand assets produced
Input from the team

"lojista na frente de sua pequena loja de roupas"

Generated by the agent

Modern editorial illustration, Brazilian shopkeeper in front of a small popular clothing store, no outlines, flat color blocks, localized grain shading only on shadows, elongated slim body, Sebrae palette: #eff3ee #65b7fb #0b2574 #f39d72 #9ff0bd. No text, no logos.

Stack: Claude Projects · GPT Image 2 · style decoding (linework, palette, composition, texture, framing).

Illustration of a popcorn vendor under a yellow ipê tree in a Brazilian street, with the message 'Aqui tem gente que faz acontecer'
Illustration of a popcorn vendor in a Brazilian town square with a family, a dog and a street musician in warm yellow and red tones Illustration of a Brazilian confectioner in a cozy sweet shop Illustration of three people working in a small sewing studio Illustration of a Brazilian entrepreneur reviewing data on a laptop
03 Sanofi · PSP Viva 2024

Vitória — an AI conversational avatar for chronic patient support.

Sanofi's Patient Support Program needed to scale 24/7 support to patients living with chronic conditions — without losing the empathy that defines the program. I led the creative direction of Vitória: a Brazilian persona designed with ChatGPT for voice and conversation, with the avatar itself generated in Recraft, plus a strict guardrail architecture and escalation paths to human specialists.

  • 10,000+conversations in month one
  • 95%CSAT in pilot
  • −40%manual support tickets
  • 15%safely escalated to humans

What I led: persona design and conversational tone (ChatGPT), avatar generation (Recraft), tone guardrails, pilot iteration with real patients, handover dashboards for the Sanofi team.

Vitória, the AI conversational avatar designed for the Sanofi PSP Viva program — a stylized 3D character of a young Brazilian nurse in a white-and-purple uniform, holding a tablet with the Programa Viva logo.
Vitória, the avatar designed for the Programa Viva
04 CBMM 2025

Turning niobium into a paid-media performance engine.

CBMM is the world leader in niobium — a critical-minerals story most audiences don't yet understand. We built a year-long paid-media architecture in Google Ads (Search, Demand Gen, Performance Max) across three creative tracks: demystifying niobium, anode batteries, and the 70-year institutional story.

  • 30.5M+impressions
  • +76%above projection
  • 5.58%CTR (2× benchmark)
  • 42.5%conversion rate

What I led: creative direction across three narrative fronts, alignment with media strategy, brand consistency between paid surfaces and CBMM's institutional positioning.

05 Valtech Latam · multi-client 2017–present

Running the Latam creative engine — across pharma, agribusiness and SMB.

Beyond AI experiments, I lead the day-to-day of Valtech's Latam creative team — the team behind AstraZeneca's PSP FazBem program in Brazil, the omnichannel program that scaled Sebrae across all 27 states, three years of full-service AOR for Zoetis in Latin America, and the global data-driven HCP engagement program for Sanofi across 26 markets.

  • 1.5B+customer interactions — Sebrae omnichannel
  • +371%registrations — Novo Nordisk NovoDia
  • +243%cumulative clicks — Sanofi global HCP
  • 3 yrsfull-service AOR — Zoetis Latam

Selected program — AstraZeneca PSP FazBem (Brazil): took monthly impressions from 665K to over 3 million, drove 5.9M video views, grew followers +11.4%, accumulated 10.3K interactions, and delivered the most engaged post in the program's history within our first month on the account.

Sectors & markets: pharma (AstraZeneca, Sanofi, Novo Nordisk, GSK, Galderma) · animal health (Zoetis) · SMB enablement (Sebrae) · industrial & materials (CBMM, Interceramic) · consumer goods (Softys, Japi). Brazil and Mexico.

Approach

How I direct AI through creative work.

  1. A

    Decode the brand before touching the model.

    Every project starts with a deep read of the brandbook and a structural analysis of reference assets: composition, palette, type, lens, framing, narrative. Without that, AI just produces louder noise.

  2. B

    Test engines like a craftsperson tests materials.

    Firefly, Flux, GPT Image 2, Nano Banana 2, Recraft, Ideogram, Claude — each one has a grain. I pick the engine for the brief, not the brief for the engine.

  3. C

    Build agents, not prompts.

    A one-off prompt is a trick. An agent — a project loaded with the brandbook, the visual references, the guardrails — is an asset. It's how the team keeps shipping after I leave the room.

  4. D

    Iterate intentionally, document everything.

    Generation in batches, comparison against the brandbook, refinement in natural language. Then a usage guide so the team can operate the agent without me.

About

Creative direction, with engineering muscle.

I'm a Creative Director based in Florianópolis, leading Valtech's Latam creative team. 20+ years in communications, 7 of them deep in AI. My background is classical — agency creative, brand systems, campaign craft — but for the last seven years my obsession has been the same question: how do you scale creative work without diluting the brand?

The answer I keep coming back to is AI agents built around the brand itself. Not generic generators. Specific systems, loaded with the brandbook, tested against the brief, owned by the team.

I've shipped this approach for Sebrae (image and illustration), Sanofi (conversational AI for patient support), and several others. The work is in production. The teams own it. That's the bar.

Get in touch

Looking for someone who can direct AI through real creative problems?

I'm open to senior creative leadership conversations, advisory work, and collaborations on brand-aligned AI systems.