Rihla — رحلة
A travel agent that knows your taste and respects your faith.
🌍 Live demo: https://rihla-agent.vercel.app · 🎬 Video (85 s): https://youtu.be/djEqkOn2J7s · 📦 Code: https://github.com/me7ko-dev/rihla ·
Rihla plans city trips for Muslim travellers and families. Tell it where you are going, who is travelling and
what you love — films, books, authors, music, games, teams — and its agent builds a day-by-day plan:
- 🎯 places that fit your taste, with an honest reason: "Because you love Orhan Pamuk", "Known for dinosaurs"
- 🍽️ halal food with an honest confidence level — Listed halal, Likely halal or Ask — and a clear
Serves alcohol flag; bars, pubs and meyhanes are never suggested
- 🕌 every day built around the prayers: Dhuhr, Asr and Maghrib at the mosque nearest to where you are at that
moment, with the day's real prayer times and Hijri date — and Jumu'ah at a well-known mosque on Fridays
- 👨👩👧 family-aware: Good for kids places, famous highlights mixed with personal finds, walking times at
children's pace
- 📱 a companion on the day: open the plan on your phone during the trip and Rihla shows what is now and
next, a countdown to the next prayer with the mosque nearest to you, and halal food, mosques and more places
around you — taken from the places Qloo already found for your plan, so it costs no new requests and works on
weak internet. Before the trip, Preview any day with a time slider.
- 🧭 one tap to Google Maps or Apple Maps for every stop and every walk; trips stay saved on your phone
Built for the Qloo Agentic Hackathon.
Why
Most AI travel planners are culturally blind: they send a Muslim family to a pub for dinner and forget that the
day has five prayers. Rihla is the opposite — it understands both what you love and how you live.
How Qloo powers it
Qloo is behind every kind of stop, not just one feature:
| What | Qloo API |
|---|
| Your favourites become taste signals ("Harry Potter" → the book, not the album) | /search, ranked by name match, type and Qloo popularity |
| Sights ranked by your taste and by Qloo audiences Islam and Parents with young children | /v2/insights with signal.interests.entities + signal.demographics.audiences |
| Honest "because you love X": each taste is asked on its own, and a place carries the badge only if that taste lifts it at least 5 places above the same query without any taste (the city's famous ten are never "because you love Frozen") | /v2/insights, one query per signal + one baseline |
| Topics you wrote ("dinosaurs", "calligraphy") become Qloo tags and rank places too; Known for dinosaurs only when Qloo tags the place with it, or reviewers' keywords are exactly your topic ("dinosaur fossils" yes, "street food" for a street-art lover no) | /v2/tags → signal.interests.tags, place keywords |
| Famous highlights next to personal finds | /v2/insights without signals |
| Halal restaurants ranked by your taste; "listed" (halal-restaurant category) vs "mentioned" (reviews) | filter.tags=urn:tag:cuisine:qloo:halal, place tags |
| No bars, pubs, nightclubs, meyhanes — excluded on Qloo's side | filter.exclude.tags |
| Mosques, photos, descriptions, good for kids, opening hours, what reviewers mention most ("dinosaurs" at the Natural History Museum) | place properties |
Never a hotel, shop, office or a place closed for good as a "sight" (rooms, check-in times, is_closed) | place properties |
OpenStreetMap adds halal-tagged restaurants and more mosques; AlAdhan gives prayer times (calculation method per country).
The agent
flowchart LR
U["Traveller: where, who, what they love"] --> A["LLM agent<br/>NVIDIA Nemotron 3 Ultra"]
A -- "taste_entities" --> Q1["Qloo /search<br/>names → taste signals"]
A -- "taste_places" --> Q2["Qloo /v2/insights<br/>signals + audiences Islam, families"]
A -- "halal_food_near" --> H["Qloo halal restaurants<br/>+ OpenStreetMap halal tags"]
A -- "mosques_near" --> M["Qloo + OpenStreetMap mosques"]
U --> P["AlAdhan prayer times"]
Q1 & Q2 & H & M & P --> W["Plan written by the agent"]
W --> C["Code checks, in rounds: real places only, opening hours,<br/>prayers at the exact time, the mosque on the way,<br/>Jumu'ah, short routes, meals open and nearby, no repeats"]
C --> E["Last model call: title, themes and tips<br/>match the final stops"]
E --> R["Day-by-day plan on a map,<br/>and a companion on the day"]
An LLM agent (NVIDIA Nemotron 3 Ultra; fallbacks Nemotron 3.5 Lightning and Groq) researches with tools, then writes the itinerary:
以上为 README 的开头部分。其余内容请阅读原文。 查看完整 README ↗
