AI built a Bangkok flood map in 3 hours; copy it in 5 steps
AI reads Thai flood reports and gives the water depth. The map first went live at 09:31 with 459 reports, and the case study shares the prompt to follow.
Correction · An earlier version of this story, published on 22 October, reported that when SV News checked the map that morning, it showed no flood reports from the previous 12 hours and its status bar showed at least three of its eight data feeds failing. A rewrite on 29 October removed that finding without telling readers, along with the note that SV News had not checked the build times and report counts SV Academy gave. Both are restored in this version.
SV Academy and SV News share a publisher.Details +
Jump to a section
Every heavy rain in Bangkok brings the same two questions: where is it flooded, and which way should I go? SV Academy used AI to build a map that answers both on one phone screen, in about three hours.
SV Academy, a Bangkok AI school, used Anthropic's Claude to build a live Bangkok flood map on the morning of 27 September 2026, a day after the Bangkok Metropolitan Administration (BMA) declared all 50 districts a disaster zone following almost 300 millimetres of rain, AFP reported.
That makes it a useful example for anyone who wants to build their own tool with AI, because SV Academy's case study shows the whole thing: the problem, the timeline and the prompt it used. After reading, you will know how the map works, what the AI did that morning, and how to make one yourself. The build times and report counts in this story come from that case study and have not been independently checked.
One map that pulls in flood reports from several sources
Flood information is spread across many places. This map pulls it onto one satellite map that opens straight away on a phone, with nothing to install.
- Traffy Fondue: flood reports from Bangkok's public complaint platform.
- Longdo: traffic incidents on the roads.
- City sensors: water levels on the roads.
- Water stations: water levels in the canals, and rainfall.
Traffy Fondue is already a channel the BMA asks residents to use. On 5 September, ahead of heavy rain forecast for 5 to 7 September, the BMA asked people to report flooding through its 1555 hotline or the Traffy Fondue app, Thairath reported.
AI reads Thai reports and works out how deep the water is
The most interesting part is how it reads the reports.
The case study explains that AI reads each Thai report and estimates whether the water is ankle, knee or waist deep, whether it is rising or falling, and how far to trust the report, filtering out junk along the way. The prompt asks for a small, cheap model to do most of this work.
Put simply, the AI acts like a helper who reads every message people send in, then drops a pin on the map showing how deep the water is there.
Besides flood spots, the map has a 24-hour timeline for each area, a route search that steers around flooded spots, and a ‘low ground’ layer that shows which areas flood first. Anyone who sees flooding can report it in four steps, with a photo or video and no sign-up.
The map itself says its data comes from the public and agencies and is not an evacuation order, and tells users to follow BMA announcements and call the 1555 hotline.
When SV News checked the map between 10:22 and 10:45 on 22 October 2026, it loaded but showed no flood reports from the previous 12 hours, and its status bar showed at least three of its eight data feeds failing, including Traffy Fondue, which had been down since 14 October.
AI built, tested and published it in one morning
According to the case study, the project owner only told the AI, in plain words, what they wanted.
SV Academy says Claude wrote the code and tests, published the site, opened the live site to check its own work, and fixed its own bugs. Here is the timeline for the morning of 27 September 2026.
- 07:57, start: SV Cloud, the hosting service the map runs on, had code storage, a database, file storage and a web address ready, and the AI began work.
- 09:16, first version: the map, the report feeds, AI reading of the text and a phone layout.
- 09:31, live: the map went online for the first time, with 459 flood reports.
- 10:26, sensors added: the city's road water sensors, updated every 10 minutes.
- 11:04, complete: flood-avoiding routes, the low-ground map, timelines and the junk filter were live, after 25 rounds of changes.
From the start of work to the first live version took 94 minutes, and the map had 600 or more flood reports within its first 12 hours, according to the case study. SV Academy announced the map on Instagram the same day.
The case study explains that the slowest part of making an app is usually not writing the code but getting it online: setting up a server, connecting a database and storing secret keys. SV Cloud is built so the AI can do those steps itself.
You can make the same kind of map in 5 steps
The case study says you do not need to know how to code. It takes about 30 minutes of your own time; the rest is waiting for the AI.
- Get 3 things ready: Claude on a Pro or Max plan, which you pay for, a free GitHub account to store the code, and a web host.
- Pick a host: this map uses SV Cloud, and the case study says the prompt also works with Vercel, Railway, Cloudflare, Render or Netlify.
- Open Claude in your project folder: in the Claude app, go to the Code tab and pick your project folder, or a new empty one.
- Paste the prompt: copy the prompt from the case study, change the first line to match your host, and send it.
- Answer its questions, then try it on your phone: Claude stops only for a few things, such as an API key, a password that lets your app use an outside service. If something is off, say so in plain words, for example "the buttons are too small".
Rules in the prompt that work for other projects too: have the AI work in small pieces and check the live site itself each time, write tests for every feature and never publish if a test fails, never put API keys in the code, and remove GPS location from photos people send in.
All the data sources in the prompt are public. If you plan to use them commercially, the case study advises reading each source's terms of use first. Then start with a question the people around you ask every day, the way this map started from two questions about floods.
FAQ
Who built the Bangkok flood map?
SV Academy, a Bangkok AI school and the education brand of Silicon Valley Innovation, used Anthropic's Claude to build and publish it on 27 September 2026, on SV Cloud.
Where does the map get its data?
The case study says it pulls reports from Traffy Fondue, Longdo, Bangkok's road water sensors, and canal water levels and rainfall. People can also report flooding through the map, and AI reads each Thai report to estimate the water depth.
Do I need to know how to code to follow it?
No, the case study says. It takes about 30 minutes of your own time, but you need Claude on a paid Pro or Max plan, a free GitHub account and a web host.
Can I rely on the map during a flood?
The map says its data comes from the public and agencies and is not an evacuation order, and it tells users to follow BMA announcements and call 1555. On the morning of 22 October 2026 it had no reports from the previous 12 hours, and several of its feeds had failed.
Sources
- SV Academy: A live Bangkok flood map, built by AI in 3 hours
- SV Academy: Bangkok live flood map
- SV Academy on Instagram: flood map announcement, 27 September 2026
- The Star (AFP): All 50 Bangkok districts declared disaster zones as flooding worsens
- Thairath: Bangkok Alerts Public to Heavy Rain Warning from 5-7 Sep
