Security and Compliance
DSO needed sovereign, multilingual large language models it could train and deploy fully on premise for defense workloads, but it lacked at-scale operational experience with the mixture-of-experts architecture it considered essential for next-generation systems.
Reported impact
The collaboration accelerated DSO's model-development timeline, delivered leading-in-class Southeast Asian language benchmark performance for operational use cases, and strengthened DSO's internal ability to build future edge reasoning models independently.
DSO used Mistral's Forge to accelerate creation of a sovereign multilingual large language model for defense operations. The program started from DSO's existing pretraining work but focused on mastering mixture-of-experts architecture, which the lab viewed as critical for next-generation LLMs. Working in a shared development environment with Mistral, the team trained a specialized MoE model on language data that included content sourced through AI Singapore, then prepared it for entirely on-prem deployment so DSO could avoid external-vendor dependency and supply-chain risk. The engagement was not only about shipping a model: Forge gave DSO repeatable training workflows while close collaboration transferred practical know-how about what works at scale. That left DSO with both a benchmark-competitive multilingual model and stronger in-house capability for later edge reasoning systems.
Language content sourced via AI Singapore feeds a shared development environment, Forge orchestrates training for the multilingual MoE model, and the resulting model is prepared for entirely on-prem deployment and later edge reasoning work on autonomous systems.
"DSO needed to train its own multilingual large language models from scratch that could be deployed entirely on premise."
"Partnering with Mistral gave us direct access to expertise in MoE architecture, helping us accelerate our model development timeline."
"The custom MoE model delivered leading performance in its class on Southeast Asian language benchmarks"
"The two teams co-developed a MoE model using Mistral's model development stack, later integrated into Forge."
"Training data included language content sourced via AI Singapore (AISG)."
Review how this reference was scored and trace each normalized claim back to its documented source.
Scoring methodology
Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.
Traceability
Field coverage shows which facts each source supports. Dates are shown only when documented by the source.
Related
DXC Sample Oct 1, 2026 INAIL As Italy's national agency for workplace health and safety, INAIL holds sensitive personal information and needed stronger defenses against growing phishing campaigns, sophisticated malware, and potential data breaches,… 11,000 nontechnical users MineMeld · classifier systems · external integration layer 
NVIDIA Sample Jul 7, 2026 French Ministry of Economy and Finance French public servants needed to answer citizen requests against more than 100 million documents per year while maintaining legal accuracy, security, confidentiality, and compliance, but manual retrieval and analysis of… 1,000 work hours saved per month for a typical 100-user deployment DeepBrain · NVIDIA AI Enterprise · NVIDIA DGX platform NVIDIA Sample Jul 7, 2026 Public Ministry of Rio Grande do Sul (MPRS) MPRS needed to manage rising legal caseloads, archived police inquiries, and public-service requests more efficiently, but the volume of legal documents and the need for compliant, locally controlled analysis made… 497 municipalities Archiving Intelligence Assistant (AIA) · Citizen Access Agent · Justice Intelligence platform 
