❝

Handwritten annotations + fuzzy maps + paper documents = a bottleneck problem in urban development? Now, AI can solve it in just 40 seconds! The British government uses Google Gemini multimodal technology to reshape the urban planning system, and 350,000 backlog applications across the country will soon be "decoded with one click"!

Introduction: AI is more than just a chatbot, it is reshaping government efficiency

In the UK, an application for house construction or renovation often takes weeks to be approved. The root cause is the large amount of urban planning documents based on paper files, with blurry drawings, sloppy annotations, and scattered information. Planners often spend  2 hours manually processing a document.

But now, this process will be completely rewritten by artificial intelligence.  The Extract system  launched  by i.AI, the AI ​​incubator team of the British government , shortens the approval time to  40 seconds by connecting to Google's multimodal large model  Gemini , and plans to  open it to all local governments across the country in the spring of 2026 !

Traditional dilemma: three "invisible barriers" behind an old drawing

1.  Unstructured information : hand-drawn maps, blurry handwriting, and embedded content that is difficult to parse

2.  Manual comparison is slow : Approver needs to repeatedly check policies, maps, and planning red lines 

3.  Limited digital transformation : no automated channel from drawings to databases

The UK has to process more than  350,000 such applications every year, which consumes huge amounts of manpower time and financial costs.

Technological breakthrough: Gemini empowers Extract, revealing the 40-second AI decoding process

❝

"It used to take 2 hours, but now it can be done in 40 seconds." - UK government official

The Extract system is built on  the Google Cloud Vertex AI platform and uses Gemini's powerful  multimodal reasoning  and  tool calling capabilities to complete the fully automatic processing flow from image to coordinates:

Visual reasoning: graphic recognition + semantic understanding

• Polygon Extraction : Identify areas in the map such as red outlines (residential areas) and gray shadows (parks)
• Integrated tool chain : call OpenCV, Ordnance Survey, Segment Anything, etc. to automatically extract geographic boundary shapes

Geolocation: Historical drawings precisely aligned with modern maps

• Identify addresses, intersections, landmarks, etc. in images
• Using LoFTR and other image matching models to achieve coordinate conversion between ancient and modern maps
• Output  coordinate data that can be directly used in GIS system (error controlled within <0.1m)

❝

Technical highlights reference: Gemini multimodal update

Pilot results: Four major cities launched a revolution to speed up approval

Extract has been piloted in four local governments, with positive feedback:

Pilot officials gave feedback: “The approval documents that used to be like archaeology can now be automatically parsed into actionable data by AI.”

National Strategy: The AI ​​Engine Behind 1.5 Million Houses

UK Prime Minister  Keir Starmer  stressed  at  London Tech Week :

❝

“We are using AI to work with businesses to drive change in the UK’s outdated approval system, which has slowed down development in the past. Extract is an important part of our ‘Plan for Change’, which aims to build 1.5 million homes and create a planning system for the 21st century.”

✊  Policy objectives:

• Reduce approval time by 90%
• Realize national digital approval
• Unlock more land resources for housing construction