Mission‑focused AI: How HTX transforms Singapore’s public safety with custom and local AI.

Key takeaways:

• State-of-the-art, public-safety-adapted models built and owned by HTX

• Competitive performance across standard LLM benchmarks

• AI models and full-stack engineering platform, deployed internally within HTX

“Our partnership with Mistral AI has accelerated the development of safer and smarter AI solutions which has greatly enhanced the productivity of public safety officers in Singapore.” – Ang Chee Wee, Chief AI Officer and Assistant Chief Executive (Digital & Enterprise), HTX.

HTX, the science and technology agency of Singapore’s Ministry of Home Affairs, is transforming public safety through AI. To succeed, HTX needed AI models that understood Singapore's unique language and context, the ability to train on data in a secure environment, and proven expertise to build state-of-the-art systems. By partnering with Mistral AI, HTX is creating a robust, locally controlled ecosystem that brings together people, products, and tooling to rapidly develop and deploy mission-focused AI capabilities. This collaboration enhances operational effectiveness, accelerates frontier R&D, and ensures data safety, enabling faster, data-driven responses and freeing up resources for public safety officers on the frontline.

Building AI that meets Singapore's needs

HTX’s mandate is to advance science and technology, working closely with public safety agencies such as the Police, Prisons, and the Immigration Checkpoints Authority etc. to augment their operations and help keep Singapore safe and secure. That mission demands AI that can handle real-time input modalities such as visual information and text, and produce textual outputs tailored across various use cases, like border operations and cybersecurity response. Generic, off-the-shelf models were never going to be enough. 

The core challenge was three-fold. HTX needed models that understood Singapore's specific linguistic and operational context. It needed to deploy those models on air gapped infrastructure with no access to web search, hence the critical domain knowledge needs to be baked into the model weights. And it needed the research and engineering depth to build and sustain world-class AI systems over time - not just deploy a vendor's product, but own the capability.

Language models are core strategic capabilities for HTX, strengthening operational effectiveness and enabling the agency to build AI tailored to Singapore’s unique needs. The right partner made all the difference. Mistral AI's collaborative approach and openness in sharing methodologies and product practices accelerated HTX’s path forward. Building specialized AI capabilities from scratch can take years. With Mistral AI's expertise, HTX established clear training, data, evaluation frameworks and structured product strategies. This enabled consistent, repeatable results in record time.

Why Mistral AI?

The decision to partner with Mistral AI came down to three things: technical excellence, genuine openness, and a shared belief in sovereign AI.

HTX required models that could run entirely on NGINE, its first enterprise-grade AI infrastructure. Data control and regulatory compliance were non-negotiable. Mistral AI’s willingness to share research, development practices, and model weights - and to support on-premises deployment - made it the ideal partner.

The partnership began in March 2025 with foundational model training and expanded quickly. The result: a structured ecosystem that combines HTX's engineers and product teams with Mistral AI's platform ( AI Forge for model training on proprietary data, and AI Studio for AI operationalization and foundational models) and embedded AI expertise - advanced AI research scientists, AI engineering capabilities, and AI practices that standardize how HTX builds, evaluates, and ships AI products.

Transforming operations, one model at a time

The first concrete output from this collaboration is HTX's Phoenix family of AI models. Phoenix Small and Medium models, pre-trained on Mistral AI’s foundation and refined on HTX's proprietary data, are now deployed on NGINE and performing at the top-tier of standard LLM benchmarks. Additional models trained extend HTX’s reach into specialized domains including embodied AI, video AI, and cybersecurity.

For public safety officers, the impact is immediate across various applications, such as:

• Intelligent decision support lets officers ingest and summarize real-time intelligence, such as incident reports, social media, and sensor feeds, enabling faster responses to fast-moving situations.

• Automated workflows handle document classification and chatbot assistance, cutting administrative load and returning time to frontline staff.

• Rapid prototyping is enabled by Mistral AI’s open-model licenses, reducing external dependencies and building a self-sufficient development cycle.

Running these custom models on-prem meets all of HTX's needs for efficiency, cybersecurity, and data confidentiality, as this partnership is about building AI that truly serves Singapore’s public safety agencies.

Defining the frontier, not following it.

HTX is not catching up to the state of the art in public safety AI. It is setting it - combining Mistral AI's research capabilities with HTX's operational context and productization discipline, creating a repeatable engine for capability development. Mistral advances foundational model performance. HTX translates that into deployable products tailored to real public safety use cases. The two reinforce each other.

The collaboration is already expanding into specialized AI for robotics, video analytics, fire safety systems, and advanced cybersecurity. Each new domain builds on the same infrastructure, the same evaluation practices, and the same shared research culture that helped build Phoenix.

HTX’s mission is to build AI that serves Singapore’s public safety agencies, and Mistral AI is a critical partner in making this a reality. HTX retains control of the models, the infrastructure, and the knowledge. And that’s the whole point.