AskBahia

The wager

AskBahia bets you can produce travel content with an agent pipeline that beats what a human editor would commission from a freelancer in 2026.

The wager is technical, but the stake is personal. I've spent years in Bahia, I own a beach house on the Península de Maraú, and the standard luxury travel coverage of the region is shallow. Condé Nast Traveler will commission one freelancer with two weeks to write about Trancoso. AskBahia has sixteen destinations, each with deep, locally-grounded, voice-matched content, in two languages, scored before it ships.

The methodology and the place are inseparable. An agentic editorial pipeline, judged on five criteria, writing in two languages, about a region I actually know. The content reads honest because the system was built to reject what doesn't.

How it's built

The architecture is an agentic editorial system with three layers: a content brain that produces structured destination content, a rendering layer that assembles it into pages, and a judge layer that decides what ships.

Each destination has its own vibe profile — a separate spec that defines how it should read. Trancoso reads different from Itacaré reads different from Morro de São Paulo. The vibe profile loads at the top of every generation pass: what to emphasize, what to avoid, which adjectives belong, which are tourist clichés.

The hub prompt produces eight editorial sections per landing page:

• Intro — what kind of trip this is
• Who is this for — ideal traveler profile
• Who should skip it — who should go elsewhere
• How to get there — realistic logistics
• Best time to visit — seasonal guidance
• How many days — minimum and maximum nights
• What makes it different — differentiator vs peers
• Comparison to alternatives — named peer comparisons

The two sections worth pausing on: who should skip it and comparison to alternatives. Most travel publications structurally cannot tell you that a place isn't for you, or that a competing destination might be the better fit. The advertising model doesn't allow it. AskBahia's prompt structure builds in honest filtering at the editorial level — the agent has explicit permission, and the structural obligation, to redirect readers when a place is wrong for them.

That single design decision is the difference between coverage that reads honest and coverage that reads sold.

Plus an FAQ block (six Q&As per page), practical warnings, and metadata fields. Every output passes through the five-criteria LLM-as-judge before it ships — specificity, voice, factual accuracy, Portuguese naturalness, structural completeness. Same judge methodology as Pico, different corpus, different voice.

The rendering layer is a twenty-component template — hero, four editorial text blocks (discover, where-to-stay, where-to-eat, what-to-do, closing-tips), three venue grids, experience showcase, stat counter, map, gallery, FAQ, testimonials, CTA. The agentic content fills the slots; the template handles presentation. Editorial and design decoupled.

Bilingual is non-negotiable at every step. Portuguese is the primary locale; English is the localized variant. The judge re-scores English on its own merits, not as a translation pass.

What's working

The pipeline has produced, scored, and published:

• Sixteen destinations — all populated, each with its own vibe profile and voice rules
• 551 places — 173 attractions, 196 hotels, 182 restaurants — every entry written by the pipeline and scored by the judge
• Six experiences — curated multi-venue itineraries, the harder editorial format
• Twenty-three landing pages — 16 destination hub pages (5 published, 11 in review) and 7 intent pages (6 published, 1 in review)

Every published page has a tagline, a description, a vibe profile, structured fields, and an SEO-ready slug. None of it is template filler. The pipeline doesn't generate at scale by lowering the bar — it generates at scale by enforcing it.

The five-criteria judge runs on every entry: specificity, voice, factual accuracy, Portuguese naturalness, structural completeness. The same discipline that runs Pico's restaurant entries runs AskBahia's destination pages — different corpus, different voice, same standard.

What's next

Two surfaces are still empty.

The articles surface — the blog component — has zero posts published. The hub and intent pages are the structural backbone of the site; long-form editorial is the next major content layer. The pipeline is designed to handle it, including the planned "Building a House in Bahia" series, but ship-quality posts haven't cleared the judge yet.

The properties surface — vacation rental listings — has the schema but no entries. That's the connective tissue with VisitMaraú (the marketplace for the peninsula); the two products share the agent-pipeline approach with different commercial intents. AskBahia is editorial discovery; VisitMaraú is the booking layer. Properties live at the seam.

After those land, the question becomes: deepen the sixteen destinations or add new ones? My instinct is to deepen. Sixteen places, written better than anyone else covers them, beats forty places written like every other travel site.

Bahia is more than Trancoso. AskBahia is the editorial map proving it.