Over the past two years, governments were under immense pressure to deliver digital public services quickly. But building a vaccine registration portal or contact tracing app isn’t straightforward. Some say there is too much red tape, while others lament that it takes a long time to build new tech tools from scratch.
Singapore’s Infocomm Media Development Authority (IMDA) hopes to fill this gap. Its Tech Acceleration Lab targets select enterprise tech companies that are working on local government projects. It helps them to build Proof of Concepts that comply with government security requirements.
The initiative also hosts a sandbox, where advanced enterprise tech solutions like DataRobot are easily accessible and can accelerate the delivery of government solutions. Representatives from IMDA’s Enterprise and Ecosystem Development division and DataRobot share how the lab helps tech companies serve the public sector in record time.
Steps to build AI
AI has been a hot-button topic within the public sector. Jakarta is enabling its smart city with AI. It is relying on video analytics to enforce social distancing, ensure transportation safety, and estimate vehicle tax revenue.
In healthcare, Singapore is using AI to predict the severity of pneumonia in patients so that clinicians can intervene in a timely manner.
But developing AI isn’t easy. Data scientists need to work closely with engineers and developers across the machine learning lifecycle to prepare and deploy models. This involves a series of steps that include data preparation, model development, model testing, model deployment, and ongoing model management.
Data scientists first need to prepare a dataset for machine learning. This process involves finding out what data is available, choosing what data to use and at what level of detail, and understanding how to combine multiple sources of information.
For example, in order to predict whether a shipment will be late, data scientists have to pick out information such as item quantity and manufacturing site. They will also have to remove irrelevant details like the price of a product.
Following this curation, software engineers have to build and train algorithms. IT teams then need to monitor if the model is operating smoothly once it is running.
Keep calm and automate everything
Enterprise tech product companies under IMDA Tech Acceleration Lab’s sandbox play a crucial role in helping government agencies to test and develop Proof of Concepts quickly. A Proof of Concept is important because it demonstrates that a proposed product is feasible before enterprise tech companies can put the real product into action.
One such company is DataRobot, which has been accredited by IMDA under its Accreditation@SGD programme. The DataRobot AI Cloud platform offers a unified environment that is built for continuous optimisation across the entire AI lifecycle.
First, DataRobot can help both novice and expert users to quickly and interactively explore, clean, and shape diverse data into AI assets ready for machine learning.
Second, DataRobot can help government agencies to develop AI models at scale and recommends the best models to implement. There is also ongoing monitoring of the models so that they can respond to new variables and the changing external environment.
The platform enables users to build dozens of models for many different use cases, such as detecting fraudulent insurance claims, predicting the likelihood of loan default, and keeping score of anti-money laundering activity.
DataRobot’s automated machine learning capabilities can help to speed up model development. “This gives government agencies the full flexibility and speed to experiment and compare the models instead of building them from scratch,” Gillian Cheng, AI Success Director of DataRobot says.
Third, after government agencies successfully deliver their product, DataRobot’s Machine Learning Operations features enable users to continuously monitor their models for performance and compliance.
For instance, DataRobot worked with a Singaporean government agency to build an email sorting system to resolve public complaints quickly. The AI solution automatically and intelligently classified the complaints received for more efficient processing, eliminating the need for manual routing.
“DataRobot’s Machine Learning Operations also automatically retrains the model if its performance deteriorates,” Cheng shares. This ensures that the model is working accurately and is continually optimised.
As mentioned previously, before enterprise tech companies roll out any products for the public, they must first prove that it works on a smaller scale. In Singapore, these companies face a unique challenge – they can only test Proof of Concepts on the Government Commercial Cloud as authorities prohibit sensitive data from leaving the country.
“The main challenges for enterprise tech companies lie in the long discovery process of government security and deployment requirements, as well as self-exploration of the setup procedure,” Fu Ting, Senior Manager at IMDA’s Enterprise and Ecosystem Development division says.
“TAL also helps to shorten the time required for government agency clients to procure and set up a fresh test environment”, she adds.
Understanding government security requirements
Despite having proven solutions, it remains a challenge for many enterprise tech companies to deploy their products because they “may not understand government IT architecture and security policies,” Fu says. “This is especially the case for some young startups who are not familiar working with government agencies,” she adds.
Singapore’s government has a specific set of standards on data integrity and security, which are not made publicly available.
As enterprise tech companies present their approaches to building a new tool, TAL will also provide technical expertise and review what can or cannot work for government agencies. For example, they will advise these companies on how they should tier their servers in the internet and intranet zones to maximise security.
It typically takes about six to nine months from POC proposal to deployment. But “with the help of TAL, enterprise tech companies may achieve this under just two months,” Fu shares.
Introducing digital public services can be a lengthy process, but it doesn’t have to be that way. IMDA’s TAL is breaking down barriers by helping companies to understand what it takes to build a tech product for government, while DataRobot is playing an active role in realising AI solutions.