Spaceship

Prototyping in the Government Today

Building prototypes and applications in the government is a time consuming and laborious process. Internal approval and deployment processes are complicated, with the estimated time taken from idea inception to deployment being upwards of 2 years.

"Time taken from idea inception to deployment being upwards of 2 years"

Most government teams do not have in-house engineers to collaborate on solutions. What happens is either the project stalls and never takes off, or more typically, an external software vendor is brought in. These vendors quotes range from tens to hundreds of thousands for a relatively straightforward prototype. The application then takes months to come into fruition due to time spent settling contract terms and getting context from the team.

"Vendors quotes range from tens to hundreds of thousands for a relatively straightforward prototype"

As officers are building prototypes for government use, data security and policy concerns are front and center. Time and effort is needed to navigate regulations and ensure applications are compliant to internal requirements.

Why prototype?

Prototyping should be a quick and easy way of validating solutions. Officers from various public agencies each have their unique set of problems, whether it’s creating a fallback passport scanner for immigration checkpoints, or a scholarship bond end-date calculator for teachers, cannot be fully solved by off-the-shelf or cookie cutter solutions.

Officers usually spend time understanding the problem space and requirements needed. However, gets bottlenecked at the building and validation stage due to competency or resource limitations.

Spaceship aims to make prototyping quick and easy again.

Text to app, in minutes

Spaceship was prototyped during OGP’s Hack for Public Good 2025 Hackathon under the guiding principles that prototyping within the government should be:

• Fast: Prototypes should be churned out in the matter of hours instead of months.

• Simple and intuitive: Public officers should easily be able to navigate around the app and prototype in their own time.

• Comprehensive: App should create functional prototypes that self-corrects in the generation process and give recommendations to cover the user’s blind spots.

The current iteration of Spaceship allows public officers to do just that. With no prior coding experience needed - just describe what you want, and a fully functional prototype is automatically generated, built and deployed on GCC (OGP).

"Just describe what you want, and a fully functional prototype is automatically generated, built and deployed"

Just like that, instead of waiting for days and months, Spaceship can get public officers from idea to workable prototype in the matter of minutes.

It’s not rocket science

Spaceship leverages OpenHands, a cutting-edge open-source AI software development system, to transform text descriptions into working applications. The system operates through a sophisticated three-phase process:

1. Understanding & Planning: When a public officer inputs their requirements, OpenHands uses advanced language models, specifically Claude 3.5 Sonnet v2, to understand the request and break it down into concrete development tasks. This process is similar to how a human software architect would plan out an application's development.

2. Code Generation & Testing: The heart of Spaceship is the CodeAct agent - an AI system that can write code, test it, and fix errors automatically. This agent operates in a secure sandbox environment, meaning it can safely experiment with code without affecting other systems. The agent can perform everything a human developer would do: writing code, running tests, and even browsing documentation when needed.

3. Iterations: If users are not satisfied with the code generated or have any modifications to make, they can simply continue conversing with the agent, which will do the heavy-lifting for them based on the 2 points above.

Agents at work

Under the hood, a component called the "Runtime" is used. This component provides a secure environment for code execution. This is crucial for government applications where security is paramount. The system also employs LLM agents - specialized AI components that handle specific tasks like managing databases, creating user interfaces, or handling API integrations.

What makes Spaceship particularly powerful is its ability to understand context and maintain consistency across the development process. When making changes or adding features, it considers the entire application architecture rather than just making isolated modifications. This results in well-structured, maintainable applications that follow software development best practices.

Through this technology, Spaceship enables public officers to rapidly prototype and iterate on applications without needing technical expertise, significantly accelerating the government's digital transformation efforts.

Each prototype developed through Spaceship is deployed on its own isolated environment within GCC (OGP). This means that the prototype source code as well as any data collected cannot be accessed except through the developer portal or the prototype application itself. Each prototype is also prevented from accessing the environments of other prototypes.

Build faster, ship instantly

Spaceship is currently in beta where public officers have been using it to build prototypes. Hear from some of our initial users below:

What’s next

Beta rollout

Spaceship is current live and have started out beta rollout. There are clear features on our roadmap that we plan to build. The priority will be informed by the usage and feedback from our beta users. Some of these features include:

• Allowing officers to easily build a wider range of LLM apps (e.g. text-to-image, text-to-video).

• Including existing products in Spaceship’s knowledge sources to allow easy building of prototypes around these products (e.g. simple wrapper around FormSG using their existing SDK).

• Permission controls and authentication for shared prototypes.

• Enabling better checkpointing of working iterations.

• Figure out optimal Spaceship prototype-to-enterprise application handoff.

LLM space

Large Language Model (LLM) systems perform increasing well on structured, coding related tasks (i.e. Anthropic's Claude 3.5 Sonnet v2 scoring 53% on the SWE-bench verified benchmark).

Recent advancements such as OpenAI's o3 model and Deepseek R1 are testament to the rate of LLM research. The industry is rapidly moving on code generation too, with commercial tools (e.g. v0 and Replit) becoming popular, increasing research in the area (many new systems on SWE-bench leaderboard weekly) and tools to facilitate AI code generation being further developed (e.g. e2b). Our bet is that the pace of these developments will not slow down anytime soon. Spaceship aims to continue leveraging on this advancements and bring its benefits to public officers in the Singapore government.

Meet the team

The prototype was developed by the following team:

Si Mun Tham | Kevin Alvarez | Jacqueline Ong | Hui Qing Lin | EnYi Leong | Bryan Ong Wen Xi