One line. Saqr puts every metro, tram, bus, boat and aircraft of the UAE on one live map. Agents turn the raw feeds into a short, grounded briefing that every sentence cites, adapted to who is watching: commuter, operator or analyst. Saqr is in Arabic and English and can read its briefing aloud.
Theme
Smart cities & mobility (closest alternative: mission-critical operations). The use case is situational awareness for transport and city operations. The multi-agent design is how it is built, not the reason it qualifies.
Problem
- Who feels it: transport operators, city and event control rooms, and the millions of daily public-transport users in Dubai and Abu Dhabi.
- What happens today: live information is split across operator apps, weather and air-quality portals, airport data and social media. Nobody gets one picture that says what is happening, where, and what to do.
- When it hurts most: on exactly the UAE's typical disruption days: dust storms, morning fog, rare but severe rain (April 2024), extreme heat, and big-event exits such as the F1 weekend, New Year's Eve or concerts.
- What it costs: slow detection and inconsistent messages. Each control room interprets data by hand, and passengers get generic alerts in one language.
Solution: how it works
| Layer | Agent | Job |
|---|---|---|
| Sensing | Transit agent | Reads the official RTA timetable (GTFS) and estimates the position and headway of every metro, tram, bus and marine trip, clearly labelled Estimated. |
| Air agent | Live aircraft over the UAE (OpenSky) and the arrival waves into DXB. | |
| Weather agent | Temperature, apparent temperature, wind, rain and visibility, plus PM10 and dust for 6 cities. | |
| Seismic agent | Earthquakes in the region (USGS). | |
| Sea agent | Ships near Jebel Ali, Port Rashid, Khalifa and Fujairah. Simulated today; AIS is planned. | |
| Reasoning | Rules agent | Deterministic thresholds turn facts into alerts: heat stress, dust, low visibility, wind, rain, arrival wave, overnight closure, disruption, crowd surge and stale data. |
| Experience | Briefing agent | Writes 4–7 sentences for the chosen audience and language. Every sentence carries its source and data age. Stale sources are left out and listed. Predictions are labelled Forecast. |
| Voice agent | Reads the briefing and new warnings aloud in Arabic or English. |
The map and panels are deterministic. The agents only add interpretation. Today the briefing agent uses a grounded template engine. The interface is ready for an Arabic LLM (for example Jais) constrained to the same cited facts.
Capability checklist
| Capability | Status | Note |
|---|---|---|
| Map of Dubai metro, tram, buses (top 140 routes) and marine lines | Working | From the RTA GTFS timetable, rebuilt from source on every publication |
| Vehicle positions and headways | Estimated | From the timetable, not GPS: RTA and ITC publish live positions only to Google Maps |
| Aircraft over the UAE | Simulated | The OpenSky client is built in, but OpenSky does not allow calls straight from a browser (no CORS). Live flights need a one-file server proxy with a 30 s cache. Until then the app shows labelled simulated traffic. |
| Weather and air quality, 6 cities | Working | Open-Meteo forecast and CAMS air quality |
| Earthquakes | Working | USGS, 2.5+ in the region |
| Alerts engine | Working | Thresholds documented in the code (heat ≥ 41/46 °C feels-like, PM10 ≥ 150/300, visibility < 1/0.5 km, wind ≥ 40 km/h, rain ≥ 2/10 mm/h, ≥ 6 aircraft on final approach to DXB) |
| Briefing in 3 audiences × 2 languages, with citations and stale-data suppression | Working | Template-grounded |
| Voice EN/AR | Working | Browser speech; quality depends on the device's voices |
| Demo scenarios: metro disruption, dust storm, dense fog, event exit surge, heavy rain | Simulated | Shown with a striped “Simulated” badge everywhere |
| Ships | Simulated | AIS feed planned (free key needed) |
| Live vehicle GPS (RTA/ITC), Abu Dhabi buses | Planned | Needs a data-sharing agreement |
| LLM briefing (Arabic) | Planned | Same grounding and citations, evaluated before release |
| “Real data only” switch and “Data” panel | Working | One click hides every simulated element; the panel explains what is live, estimated and simulated |
| Accounts | Not needed | Private local profile only (language, audience, layers); "Clear my data" button |
Data sources and permissions
| Source | Use | Terms |
|---|---|---|
| RTA GTFS via Dubai Pulse / Mobility Database mdb-3359 | Timetable, lines, stops | Dubai Pulse open data licence (attribution) |
| OpenSky Network (planned, via proxy) | Aircraft states | Free for research and non-commercial use; a commercial pilot needs a licence |
| Open-Meteo (incl. CAMS air quality) | Weather, PM10, dust | CC BY 4.0; commercial API plan for production |
| USGS | Earthquakes | Public domain |
| OpenFreeMap (fallback: CARTO) | Base map (dark and street styles, labels in Arabic and English) | Open, no key · © OpenMapTiles © OpenStreetMap contributors |
The prototype redistributes no third-party data beyond what each licence allows. A production pilot would need agreements with RTA/ITC (live GPS) and commercial plans for flights and the basemap.
How to test it (3 minutes)
- Open the demo. The clock shows Gulf time. Around 800 vehicles move on their timetable positions. Weather, air quality and earthquakes are live. Aircraft are simulated.
- Change Audience (Commuter, Operator, Analyst): the same facts are explained differently. Switch to عربي: the layout turns right-to-left and the briefing changes language.
- Press Metro disruption: trains are held on a Red Line segment, an alert appears and the briefing recommends a bus bridge (operator) or extra travel time (commuter).
- Try Dust storm, Dense fog, Event exit surge and Heavy rain. Each one is labelled Simulated.
- Move the time slider to 08:00 or 23:30, or play at 10× or 60×, to see rush hour or the overnight closure. Press Read aloud.
Deep links: morning rush + disruption · Arabic + dust storm · fog, analyst view · event exit
Impact: pilot plan (targets, not results)
< 2 mintime from a source event to an operator-visible alert (p95), vs. a manual monitoring baseline measured in the pilot
< 2 %briefing sentences with a factual error, on 200 annotated situations
≥ 4 / 5Arabic clarity, rated by 10 native speakers
≥ 70 %of commuters who say they would change their trip, out of 30 testers
Method: a 4-week shadow pilot in one control room (8 operators) and 30 commuters. Saqr is compared with the current tools on the same days. We log alert latency, factual errors (two reviewers), relevance ratings and end-to-end data age. Every figure above is a target to measure; nothing has been measured yet.
Safety, privacy and trust
- No accounts and no tracking. Settings stay in the browser and can be cleared with one click.
- Every briefing sentence cites its sources and data age. Stale sources are suppressed and listed.
- Estimated, simulated and forecast information is always labelled as such.
- Advisory only: Saqr does not replace the operators' safety systems or official warnings (NCM, NCEMA, RTA, ITC).
Route to value for Presight
- Initial customer hypothesis: an event-mobility command centre (for example the Abu Dhabi Grand Prix week on Yas Island, or New Year's Eve in Dubai). It is bounded in time, has high visibility and a clear owner. It would then extend to RTA/ITC operations.
- Why Presight: it already has the real-time geospatial analytics and the Arabic AI to take Saqr from estimated to live data. Saqr is the citizen- and operator-facing explanation layer on top of that analytics.
- Business model: a B2G platform licence per command centre, plus an optional white-label commuter feed for the operator's own app.
Status, milestones and support needed
- Now (TRL 4): a working prototype with real timetable, flight, weather, air-quality and seismic data, and simulated scenarios. One developer; built in a day on a reusable architecture: a static web app, a build script that rebuilds the transit data from the official source, and pluggable agents.
- Next 8 weeks: live flights through a cached proxy, AIS ships, an LLM briefing with grounding evaluation, and an Arabic evaluation with native reviewers.
- Support needed: introductions for RTA/ITC data access, Arabic AI evaluation, and a pilot owner for one event.
Limits (said plainly)
- Transit positions are timetable estimates.
- The ships and all the scenarios are simulated.
- Flights are simulated until the proxy is added; after that they depend on OpenSky's free quota.
- The briefing is template-based today.
- Station names come from the GTFS in English.
بالعربية باختصار: «صقر» رادار مباشر للتنقل في الإمارات. يجمع المترو والترام والحافلات والنقل البحري (مواقع تقديرية من الجدول الرسمي لهيئة الطرق والمواصلات)، والطائرات (محاكاة حاليًا)، والطقس وجودة الهواء والزلازل في خريطة واحدة. يكتب وكيل الموجز ملخصًا موثّق المصادر للراكب أو المشغّل أو المحلّل بالعربية والإنجليزية، ويقرؤه بصوت مسموع. كل ما هو تقديري أو محاكاة أو توقّع موسوم بوضوح.