SolvTrav

Travel inventory in Egypt is scattered across listing sites, agency spreadsheets and WhatsApp threads. A traveller looking for a chalet on the North Coast for six people under a budget has to hold the requirements in their head and check every source by hand.

Natural-language search is easy to demonstrate and hard to ship. The model has to turn loose phrasing into structured constraints — area, dates, party size, price ceiling — and then answer only from inventory that actually exists and is actually available. A travel assistant that invents a villa is worse than no assistant.

A unified inventory layer underneath five verticals, with two ways into it: a conventional filtered search for people who know what they want, and a booking assistant for people who would rather describe it. Both read the same inventory, so neither can offer something the other cannot book.

• One inventory model spanning stays, hotels, tours, transport and activities
• Natural-language query understanding that resolves to structured filters
• A conversational booking assistant with saved chat history and guided fallback
• Arabic and English throughout, including the assistant
• Listing verification, so a result carries a provenance signal rather than a rating alone

• Five inventory types answer one query instead of five separate searches.
• A traveller can state a constraint set in a sentence — area, party size, budget — and get bookable results, with a guided search underneath for when the sentence is not enough.
• The assistant keeps a thread, so an interrupted search resumes instead of restarting.