Huawei Shows Off OneMap GPT Challenge Winner
- By Paul Mah
- March 06, 2024
Huawei today announced that its Huawei OneMap GPT prototype was one of two winners of the inaugural Singapore Land Authority (SLA)’s OneMap GPT Challenge.
The winning prototypes were unveiled this morning by SLA, the national geospatial and mapping agency at the Geo Connect Asia 2024 conference.
The OneMap GPT Challenge was launched by SLA in October 2023 with the goal of exploring ways to incorporate AI technologies for innovative solutions for OneMap, the authoritative map of Singapore, and to extend the benefits of geospatial to a wider community.
There’s an AI for that
Huawei’s winning prototype was developed by the Applied Intelligence Centre of Excellence of Huawei International on Huawei Cloud, and was chosen out of 41 entries.
Huawei OneMap GPT integrates geospatial and generative AI technologies with public (OneMap) and private sector data. It offers a Google Map-like interface but with a ChatGPT-like chat box instead of a search box.
Users can key in their queries to plan their daily activities or seek responses to guide decisions around their itinerary or even identify the ideal location for a home.
In a demonstration shown on stage, a speaker walked through the process of narrowing down suitable locations for purchasing an apartment based on parameters such as budget, proximity to parents, and nearby amenities. These parameters were fed in piecemeal through the chat interface.
The prototype builds on initial research and development work by Dr Ashley Fernandez, the chief data and AI officer at Huawei International and his team. When the Challenge was launched, a team of five, including Fernandez, embarked on developing it.
In an interview with CDOTrends, Fernandez explained that the initial research revolved around amalgamating disparate data types, including geospatial data. The team eventually developed a solution that included the use of a graph database to define relationships.
The hardest part was the incorporation of policies, he noted, which can be complex and written in natural language. This might include government policies that define the conditions under which citizens are entitled to rebates or subsidies when buying public housing.
“One of the underlying complexities in bridging geospatial use cases and GPT AI models goes back to data. Geospatial data by design comprises of various formats, along with its dynamic and ever-changing representation of our world in time and space,” explained Fernandez.
“It’s always exciting to reimagine the entire data and GPT stack through our extensive R&D experience, reengineering and innovating new solutions to old problems.”
Image credit: Paul Mah
Paul Mah
Paul Mah is the editor of DSAITrends, where he report on the latest developments in data science and AI. A former system administrator, programmer, and IT lecturer, he enjoys writing both code and prose.