- Home
- Case studies
- Recent Visits
Guided Labs product · Travel
Recent Visits
How we combined researched Nepal travel guides, an AI assistant that plans from them, and quotes from verified local agencies — with code, not the model, deciding what gets recommended.
- Services
- SaaS development, AI development
- Technologies
- Next.js, React, TypeScript, Supabase (Postgres, pgvector, Auth, Storage), Vercel
- Product
- Recent Visits

Outcomes
Up to 3
local agencies quote each trip
Travellers compare structured quotes and talk to agencies in one place.
Always labelled
paid Featured placement
Paying never lets a tour that doesn't fit the trip in.
Source + date
on every time-sensitive guide
Visas, permits and safety facts show where they came from and when they were last checked.
The challenge
Travellers planning a Nepal trip piece information together from many sources, and the details that matter most — visas, permits, seasons, altitude — change. A generic AI chatbot makes this worse: it will confidently invent a price, an agency or a permit rule.
We wanted a planner that only speaks from researched content and real partner offers, and a business model that never lets payment decide what a traveller is told.
Our approach
Editorial content first
Recent Visits publishes destination guides, treks with altitude, difficulty, permits and day-by-day routes, things to do, practical travel guides and suggested itineraries. Time-sensitive facts carry their source and the date they were last checked, and the site shows when a guide is due for review.
Let code decide, let the model explain
- The assistant answers through typed tools that search destinations, treks, activities, travel guides, itineraries and partner tours — not from the model's memory.
- Retrieval combines vector search with keyword search over the editorial content, preferring Recent Visits' own guides.
- Which partner tours to show is decided by a deterministic recommendation engine that filters for fit, scores, and returns its reasons; the model only narrates them.
- Day-by-day outlines come from a code planner or a published itinerary, never from free-form generation.
- The assistant is instructed never to invent agency names, prices, availability, visa or permit rules, to label Featured results, to treat retrieved text as data rather than instructions, and to stay on Nepal travel.
Transparent partners
Agencies pay a flat fee to be listed and receive quote requests; the fee doesn't change where a tour appears. Paid Featured placement only reorders tours that already fit the traveller's trip, and is always labelled.
The solution
Recent Visits is live at recent-visit.com. Travellers can read guides, ask Recent Visits AI to turn their dates, group and interests into a day-by-day outline, save and share trips, and ask up to three local agencies for a quote — then compare structured quotes and message agencies in their account.
Traveller stories let people share trips in their own words: they are reviewed before publishing, photos are re-encoded to remove location data, and any AI writing help is disclosed. Ads appear only on editorial pages, only after consent. The How we make money page explains every revenue source.