What tech stack does Helsing use?

Rust is Helsing's primary systems language, chosen for memory safety, deterministic performance on constrained embedded hardware, and suitability for real-time sensor processing in GPS-denied environments. Public repositories confirm the depth of the commitment: Sguaba is a Rust crate providing strongly-typed coordinate system transformations for flight and navigation; lsmlite-rs is an LSM-tree storage engine with a 16 MiB memory footprint designed for real-time sensor data on embedded devices; and Twurst is a Rust implementation bridging gRPC and Twirp for brownfield networks that lack HTTP/2 support — a critical capability in contested battlefield radio environments where standard internet protocols are unavailable.

Helsinng's infrastructure philosophy is formalized in its 'Node Computing' paradigm, introduced by CTO Dr. Robert Fink on the engineering blog. Node Computing treats battlefield systems as a distributed asynchronous network of compute nodes — each with local data and local compute, producing emergent global behavior — rather than a client-server architecture. This is a foundational difference from conventional cloud-native enterprise software and explains why Helsing builds most of its own tooling rather than adopting off-the-shelf solutions. The build and deployment chain reflects this security posture: Helsing bootstraps Nix entirely from source from a 256-byte seed binary — ensuring no unverified binary enters the build pipeline, a requirement for defense-grade software supply chain security.

For developer community and talent, Helsing is a gold sponsor of IC Hack at Imperial College London (600+ participants in 2025), featuring Rust-focused technical talks — a signal of active community investment in Rust adoption and a pipeline for Rust-specialist recruiting.

On the ML and AI layer, Python with PyTorch forms the standard research and training stack, configured via Hydra and OmegaConf for hyperparameter management — common patterns in production ML engineering organizations. Helsing's published work on vetting third-party software packages uses DistilBERT and LLMs for automated analysis, confirming active use of transformer-based models internally beyond just the weapons systems themselves.

Data interchange across distributed nodes uses Protocol Buffers managed by Buffrs — an internal package manager that Helsing open-sourced to solve distribution and dependency challenges for protobuf schemas in large polyglot codebases where multiple teams produce and consume shared schemas. For peer-to-peer resilience in mesh network scenarios where a central server is unavailable (a common reality in contested battlespace), Helsing developed DSON, a delta-state CRDT for JSON-like structures — an eventually-consistent distributed state approach that allows disconnected nodes to synchronize when connectivity resumes. This is sophisticated distributed-systems engineering well above what most enterprise software companies produce.

Helsinng's GTM tooling is not publicly disclosed. As a B2G company with a small number of high-value sovereign customers — Germany, Ukraine, Estonia, Japan as of mid-2026 — it likely uses a lightweight CRM rather than a complex sales engagement platform. No public signals on marketing automation or outbound tooling are available, and given the security sensitivity of its customer relationships, Helsing is unlikely to use standard cloud CRM with broad data-sharing.

Helsing's deeply custom, Rust-first, edge-native architecture creates specific adjacencies and displacement patterns for vendors. Cloud providers with strong edge, confidential computing, and sovereign cloud offerings (AWS Outposts, Azure Government, European sovereign cloud instances) are well-positioned — Helsing's Node Computing model requires compute at the edge, not just centralized inference, and its security requirements push toward air-gapped or certified cloud environments. MLOps platforms supporting on-premise or air-gapped deployments (not just SaaS MLOps) are more relevant than standard cloud-native tools like Weights & Biases.

Helsinng's investment in building Buffrs (protobuf package manager) and Twurst (gRPC middleware) is a strong signal of a build-over-buy posture for core infrastructure — vendors pitching commodity DevOps or API management tools will face a 'we built that ourselves' objection. The areas where Helsing is clearly a buyer are: embedded hardware (GPUs and FPGAs for edge AI inference in constrained drone compute), synthetic data and simulation environments for AI training (critical for weapons systems that cannot be trained purely in live combat), safety-critical software testing and formal verification tooling, and aerospace-grade manufacturing equipment as production ramps at RF-1 (Germany) and Plymouth (UK). The $1.2B Series E specifically funds production ramp and engineering headcount — making the mid-2026 to 2027 window a favorable buying cycle for hardware-adjacent and manufacturing-adjacent vendors.