Working demo · October 2026 · Synthetic data
An AI command centre for Pattaya City. It reads the city's records, residents' photos and Thai official letters, computes with tools and drafts the next step for a named official to approve. This is the working demo, on synthetic city data.
Open the command centreSee the four workflows
Demo sign-in: pick a role, code 1234.
Anything not yet confirmed is marked “To be confirmed”.
The city
Pattaya runs its own hospital, roads, drainage, waste collection, beaches and back office. Each service keeps its own records. The AI opportunity is to read across them, so officials see one picture and act earlier.
General public information. Figures and baselines will come from Pattaya City departments during discovery.
01 · The issue
The issues listed under each area are working hypotheses to test with the departments that own them, not findings.
In this demo
AI opportunity. Photo reports matched to a dog registry, hotspot maps and planned CNVR rounds.
In this demo
AI opportunity. A Thai-language document assistant, request triage and routing, and reports that draft themselves.
In this demo
AI opportunity. One live map of the city with what-if scenarios for flooding, traffic congestion and crowd safety.
This demo has the rain what-if only, as an illustrative model. Traffic and crowds: to be confirmed.
Not in this demo: next version, to be confirmed
AI opportunity. Arrival forecasts, multilingual intake and drafted discharge and referral documents.
Not in this demo: next version, to be confirmed
AI opportunity. Waste forecasts with re-planned routes, water-quality alerts and an AI-drafted monthly sustainability report.
Possible further use cases. Scope, owner and data are not confirmed yet.
02 · The AI
Each workflow starts from something that already happens in the city: a report, a forecast, a letter. Models do the numbers, an AI agent joins the pieces and drafts the action, and a named official approves it.
The console opens on what changed overnight, what is at risk and what to decide today, across requests, stray dogs and rain.
Requests, dog reports, bite incidents and the real rain forecast for Pattaya are read when the console opens.
Database counts, dog-report clustering, a deadline-risk model and the drainage model each raise a signal with its facts and its reason.
The agent writes the briefing from the tool output and drafts an action for each signal that needs a decision.
The duty officer decides on each draft; executives see the same picture.
Each decision is logged with who made it. Nothing is sent anywhere without approval.
Tools the AI agent calls: morning_briefingcity_overviewquery_requestsdraft_action
A resident report becomes a matched, prioritised case and a planned CNVR visit.
A resident sends a photo and taps the map in a LINE-style chat, in Thai or English (simulated, not the city's real LINE account).
The vision model counts the dogs and reads visible marks: coat, collar, ear notch. It never claims to know which dog it is.
Registry dogs nearby become candidates, repeat reports merge into one case, and bites, nearby schools and unsterilised dogs set the priority.
The duty officer approves the drafted visit and the case joins the next CNVR round.
The field team records the visit, the registry updates and the resident gets a reply in their language.
Tools the AI agent calls: dog_hotspotsdog_casespropose_rounddraft_action
A scanned Thai official letter is read, routed to the right division and answered in the official format.
An officer uploads a scanned memo or external letter, or a PDF.
The model reads the number, date, sender, subject, place, request type and urgency. Officers can correct any field.
The request type sets the division and the deadline, the place is found on the road map, similar past requests are shown, and the AI writes only the body inside a fixed letterhead.
The division head checks the routing and signs. Signing registers the request with its deadline.
The deadline board shows which requests are likely to miss their deadline this week, from a model over past requests.
Tools the AI agent calls: letters_inboxrequests_at_riskquery_requestsdraft_action
A real rain forecast runs through an illustrative drainage model on real roads and becomes a street-level plan.
The real Open-Meteo forecast for Pattaya gives the wettest 3 hours ahead; a slider asks “what if more falls?”.
Satellite terrain and 3,443 OpenStreetMap road stretches show where water collects. Illustrative, not calibrated.
The agent drafts pumps, closures and drain-clearing crews, and a Thai and English public notice that gives no depths.
The duty officer adjusts the plan and approves it.
Crews are in place before the wettest hours, and the notice goes out once approved.
Tools the AI agent calls: rain_forecastsimulate_drainagedraft_actiondraft_notice
Platform
The use cases share one data layer and one set of AI services, so the second use case costs far less than the first. Data flows up; approvals and audit wrap every step.
In this demo Command centre, division head signing, a field team page and a resident chat.
In this demo Clustering, a deadline-risk model, a drainage model and round planning compute the numbers; the AI agent reads, explains and drafts in Thai and English.
In this demo One PostGIS database: geo index, Thai text, role-based access and an append-only audit log.
In this demo Real: OpenStreetMap roads and schools, the Open-Meteo rain forecast, Copernicus elevation. Synthetic: requests, dogs, letters and residents.
03 · The screens
The opening screen is a briefing, not an empty dashboard: what changed overnight, what is at risk and what to do next. Ask a question in Thai or English, and the answer comes back as tables, maps and drafted actions.
City Hall control room, on shift. Morning briefing, map and assistant; approves drafted actions.
Mayor and senior management. Reads the briefing and asks the assistant; duty officers approve actions.
Director of a city department, e.g. Public Works. Signs replies to official letters.
Stray-dog team B on the street (CNVR: catch, neuter, vaccinate, return). Visits stops in order, records each visit.
Member of the public. Reports stray dogs with a photo, as on the city's LINE account.
Hospital team: next version, to be confirmed. Decision support. Officials approve every action.
04 · The path
This is the working demo. Pilot scope, data, budget and timeline are agreed only after Pattaya City has seen it.
Show the idea working on synthetic data, so officials can react to something concrete.
The demo runs on a Groovy Web server, not on city or Thai cloud infrastructure.
One or two use cases on real data with one or two departments, running alongside today's process.
Duration, budget and team: to be confirmed.
A city platform: more use cases on one shared data and AI layer, hosted in Thailand.
Procurement path: to be confirmed.
| Use case | Data readiness (assumed) | Delivery risk | Visible to residents | Suggestion |
|---|---|---|---|---|
| Stray dog management | Medium | Low | High | Pilot candidate |
| Internal operations | TBC | Low | Medium | Pilot candidate |
| Go Green, waste routes | Medium | Low | Medium | Second wave |
| Municipal hospital | Needs system access | Higher, clinical | High | Second wave |
| City digital twin | Needs GIS and drainage | Medium | Medium | Production phase |
A first read, for discussion with Pattaya City. It will change once real data is available.
Measures to agree for the pilotRequest resolution time Baseline TBCOverdue requests per month Baseline TBCDogs sterilised and vaccinated Baseline TBCER wait to triage Baseline TBCOverflowing-bin complaints Baseline TBC
Built for a city government
A public body will ask who decided, on what evidence and where the data lives. The design answers that from day one, and the list below is what we need from Pattaya City before a pilot.
Dispatching crews, spending money or messaging the public always waits for a named official's approval.
0 decisions by named demo roles so far, each in the log.
Numbers link back to the record or document they came from, with the inputs and the rule behind each signal one click away.
Who asked what, which data was used, what the AI suggested and who approved it, all kept on record.
48 AI and human actions logged; hash chain checked just now: intact.
Personal data is minimised and access is role-based, in line with Thailand's Personal Data Protection Act B.E. 2562. The demo holds no real personal data.
Reads and drafts Thai official documents. The demo works in Thai and English; a pilot can add the languages visitors use.
This demo runs on a Groovy Web server. A pilot runs on city servers or a Thai cloud, and the AI model sits behind one adapter, so it can be swapped for an in-country model.
Nothing below is assumed. Each item is a question for discovery.