By info
•
Jun 25, 2026
•
About Ask Santorini AI
The travel industry is broken for creators and independent builders.
For years, travel bloggers, local operators, and digital creators have relied on static guidebooks, fragmented affiliate links, and generic search engines to share recommendations and monetize their expertise. Visitors come to a page, read a blog post, and then leave to open ten different tabs on TripAdvisor or Booking.com to actually plan their trip. The creator loses the engagement, and more importantly, they lose the transaction value.
We looked at Santorini — one of the most heavily visited, high-intent travel destinations in the world — and asked a simple question: what if travel inspiration and the act of booking didn't have to live on separate pages?
That's why we built AskSantorini.ai. It's not just an AI travel assistant; it's a conversational infrastructure designed to change how destination intelligence and affiliate commerce interact — built deliberately as a reusable blueprint, not a one-off.
Moving Past the "Chat Wrapper" Hype
Let's be honest: the internet is flooded with generic wrappers that just bounce prompts to an LLM. If you ask a standard chatbot for a restaurant recommendation in Oia, it serves you frozen training data from two years ago. In a fast-moving destination like Santorini, where venues close, menus change, and new boat rentals launch every season, stale data is useless.
We built AskSantorini.ai with a source-grounded architecture.
By pairing Gemini 2.5 Flash with live Google Search Grounding and the Google Places API, our concierge doesn't guess — it checks. When a user asks about a specific venue, the system pulls structured Places data: live operating hours, phone, website, geocoding. The result is the warm, un-gimmicky persona of a local insider with the precision of a live database.
And here's the part we're most proud of, because it's the part most products skip: the bot knows the difference between what it can verify and what it can't. Contact details — a phone number, a website, an address — are only ever stated when they come from verified venue data. When a recommendation comes from general web grounding (which is great for what a place is known for but unreliable for its exact phone number), the concierge names the place and tells you to confirm the details from the official source, rather than inventing a number that might be wrong. We enforce this as a hard architectural rule across every response path, not a hopeful instruction in a prompt.
For a concierge, that discipline matters more than any feature. A travel assistant that confidently gives you the wrong restaurant's phone number is worse than one that admits it doesn't have it. We chose honesty by construction.
The Engine for Creators: Turning Intent Into a Trackable Pipeline
For creators, the real shift isn't the chat experience — it's the transactional backend behind it.
Traditional affiliate marketing means pasting clunky links and hoping for the best. AskSantorini.ai introduces an intent-aware affiliate engine running entirely at the edge.
When a user expresses intent — say, asking about private caldera boat cruises — the system doesn't just list operators. It surfaces a relevant partner offer with a discount reservation flow, right inside the chat. The backend mints an enumeration-resistant, crypto-random booking code, registers it in a Cloudflare D1 ledger, and fires a confirmation email through a serverless pipeline to both the user and the local partner.
[User Expresses Intent]
│
└──> [Contextual Partner Match]
│
└──> [Edge Engine Mints Crypto-Random Discount Code]
├──> Stored in D1 Ledger
└──> Triggers Serverless Email (User + Partner)
This is designed to capture user intent at the exact moment of inspiration and create a direct, trackable attribution loop — before the visitor leaves the page. To be precise about what that loop proves: a redeemed code is first-party evidence of intent and activation, captured independently of any third-party affiliate network. It's the signal that a user chose to act — the start of a transaction the partner can then close, not a guarantee of a completed sale. That distinction is the point: it's evidence a creator actually owns, instead of hoping a marketplace attributes a click correctly weeks later.
We're an early-stage pilot, so we're not going to wave around conversion numbers we haven't earned yet. What we can show is the mechanism, working end-to-end in production today.
Built on Solo-Founder, Zero-Host Infrastructure
As creators and indie hackers, we don't have the luxury of managing bloated dev-ops teams or paying massive baseline server bills. We designed AskSantorini.ai to scale to zero.
The entire system runs on a fully serverless, edge-native stack:
Compute: Decoupled Cloudflare Workers separating static assets from the core API, locked down with production CORS enforcement.
Database: Cloudflare D1 (SQLite at the edge) for the booking ledger, and KV for lightning-fast configuration reads.
Ops Layer: A Google Sheet acts as the single source of truth for affiliate management. A non-technical partner edits a spreadsheet row, and an Apps Script trigger automatically syncs it to the edge KV store. No backend portal to build, no servers to patch.
That architectural discipline let us solve real problems the right way — for example, handling high-frequency rate-limiting with atomic edge bindings instead of eventually-consistent KV counters (we tried the KV approach first; it didn't hold up under load, so we moved to native bindings). Costs stay near zero during low-traffic periods because there's nothing idling.
The Turnkey Vision
We didn't build AskSantorini.ai as a one-off. We built it as a blueprint.
The architecture deliberately separates the destination-specific layer — local knowledge base, persona prompts, partner spreadsheet, visual theme tokens — from the core engine. The goal is a reusable platform where standing up a new destination means swapping the config, not rewriting the code. AskSingapore.ai is our next planned instance, built to prove exactly that portability on the same shared engine.
We're honest about where we are: this is a maintained, evolving codebase with a real roadmap, not a finished monolith. That's a feature, not an apology — it means what ships is something we actually understand line by line, and what's next is on a list, not a surprise.
AskSantorini.ai is a model for creators because it shows that solo builders can use edge compute and grounded LLMs to challenge monolithic travel platforms — cleanly, affordably, and with the kind of factual discipline that earns a user's trust the second time they ask.
The era of static travel planning is ending. The era of context-aware, transactional, honest edge commerce is here.
