Defense
Proximity
Off the LLM map (in-house non-LLM ML)
Models
—
Cloud · datastore
HybridProprietary internal telemetry stores (Lattice platform)Data lake/warehouse for long-term sensor/log storage (likely cloud + on-prem, exact vendors undisclosed)Streaming/message bus layer (Kafka/Pulsar-like per system design guidance, not confirmed as production)
Compliance
ITAR (International Traffic in Arms Regulations) – required for US defense exports and autonomous weapons systemsDoD cybersecurity controls (e.g., NIST SP 800-53 / FedRAMP-aligned baselines for handling defense telemetry)SOC 2–style controls for hyperscale manufacturing and software-defined facilities (inferred from defense SaaS norms)Defense-specific secure manufacturing and export controls for Arsenal-1 and international facilities
Architecture
Anduril’s production architecture is built around its proprietary **Lattice** autonomy platform as the software backbone for autonomous systems and the **Arsenal** software-defined manufacturing stack, integrating sensor fusion, decision logic, and production control across a mix of cloud and tightly controlled on‑prem deployments.[10][12][13] Public material focuses on capabilities and manufacturing rather than concrete cloud/database vendor disclosures, so specific underlying services (e.g., exact clouds and DB engines) remain intentionally opaque.[3][10][12]
Defense
Proximity
Direct API
Models
Mistral
Cloud · datastore
HybridPostgreSQL (inferred from typical modern Python/C++ stacks)[4]Time-series/telemetry stores (e.g., proprietary or specialized DBs for sensor data, inferred)[4]On-premise data lakes for simulation/operational data (inferred)[4]
Compliance
Defense-grade secure cloud and sovereign environments (non-public details)[4]Likely compliance with European defense and export-control regimes (inferred from role as 'Europe’s largest defence technology company')[3][6]
Architecture
Helsing operates a hybrid of secure cloud and sovereign/on‑prem defense environments built around Python/C++/Rust, PyTorch, and Kubernetes for AI workloads and telemetry/simulation data processing, integrated into physical Resilience Factories for production of autonomous and EW systems.[1][3][4] Their core software stack ingests multi‑sensor and weapons‑system data to provide real‑time battlefield insights and decision support rather than public, consumer-facing LLM services.[5][6]