<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>Limina — Deep Tech Product Design</title><description>Systems engineering, research, and design for mission-critical software.</description><link>https://limina.agency/</link><language>en</language><item><title>Why Offline-First Is an Architecture, Not a Feature</title><link>https://limina.agency/research/01-offline-first-architecture/</link><guid isPermaLink="true">https://limina.agency/research/01-offline-first-architecture/</guid><description>Most software treats offline capability as a degraded mode. We argue it should be the default — and show how designing for zero infrastructure from day one produces fundamentally different, more resilient systems.</description><pubDate>Mon, 06 Apr 2026 00:00:00 GMT</pubDate><category>Offline-First</category><category>P2P</category><category>CRDT</category><category>Distributed Systems</category><category>Resilience</category><author>Anton Krupko</author></item><item><title>Terramechanics in Real-Time: From Apollo Lunar Rovers to Emergency Routing</title><link>https://limina.agency/research/02-terramechanics-in-real-time/</link><guid isPermaLink="true">https://limina.agency/research/02-terramechanics-in-real-time/</guid><description>How soil mechanics models developed for NASA&apos;s Lunar Roving Vehicle program became the foundation of a real-time terrain traversability system for disaster response — and why physics beats pixels for terrain classification.</description><pubDate>Mon, 06 Apr 2026 00:00:00 GMT</pubDate><category>Terramechanics</category><category>Bekker-Wong</category><category>GPU Compute</category><category>Machine Learning</category><category>Pathfinding</category><category>wgpu</category><author>Anton Krupko</author></item><item><title>Five Prototypes Failed Before We Got the Architecture Right</title><link>https://limina.agency/research/03-five-prototypes-failed/</link><guid isPermaLink="true">https://limina.agency/research/03-five-prototypes-failed/</guid><description>The honest story of building ARES — a P2P command-and-control platform. Five months of research, one month of Python, five structural failures, and the design principles that emerged from each one.</description><pubDate>Mon, 06 Apr 2026 00:00:00 GMT</pubDate><category>Rust</category><category>Python</category><category>Architecture</category><category>Prototyping</category><category>Lessons Learned</category><category>P2P</category><author>Anton Krupko</author></item><item><title>Polymorphic Mesh: Running LoRa, MANET, Starlink, and IP Simultaneously</title><link>https://limina.agency/research/04-polymorphic-mesh-networking/</link><guid isPermaLink="true">https://limina.agency/research/04-polymorphic-mesh-networking/</guid><description>Most mesh networks pick one transport. ARES runs them all at once — LoRa for constrained radio, QUIC for high-bandwidth, with automatic link promotion between them. Here&apos;s how polymorphic networking works and why it matters for field operations.</description><pubDate>Mon, 06 Apr 2026 00:00:00 GMT</pubDate><category>Mesh Networking</category><category>LoRa</category><category>Meshtastic</category><category>P2P</category><category>QUIC</category><category>Iroh</category><category>Radio</category><author>Anton Krupko</author></item><item><title>Cloud as Transport, Data as Yours</title><link>https://limina.agency/research/05-cloud-as-transport-data-as-yours/</link><guid isPermaLink="true">https://limina.agency/research/05-cloud-as-transport-data-as-yours/</guid><description>What happens when you treat cloud infrastructure as a relay layer instead of a custody layer? You get P2P apps where users own their data structurally — not by policy, but by cryptographic design.</description><pubDate>Mon, 06 Apr 2026 00:00:00 GMT</pubDate><category>P2P</category><category>Kora</category><category>Privacy</category><category>Encryption</category><category>CRDT</category><category>Decentralization</category><author>Anton Krupko</author></item><item><title>Semantic Voxel Encoding: Packing Terrain Intelligence into 16 Bits</title><link>https://limina.agency/research/06-semantic-voxel-encoding/</link><guid isPermaLink="true">https://limina.agency/research/06-semantic-voxel-encoding/</guid><description>How we designed a 16-bit voxel format that encodes material class, severity, elevation, road presence, lethality, and data provenance — giving pathfinding algorithms everything they need in 8 kilobytes per tile.</description><pubDate>Mon, 06 Apr 2026 00:00:00 GMT</pubDate><category>SSVG</category><category>Voxel</category><category>GPU Compute</category><category>Pathfinding</category><category>Data Structures</category><category>wgpu</category><author>Anton Krupko</author></item><item><title>GPU Wave Propagation for Pathfinding: Why FIM* Beats A*</title><link>https://limina.agency/research/07-gpu-wave-propagation-pathfinding/</link><guid isPermaLink="true">https://limina.agency/research/07-gpu-wave-propagation-pathfinding/</guid><description>A* builds graphs. FIM* solves physics. How a wave propagation algorithm from computational physics replaced graph-based pathfinding for unstructured disaster terrain — and why it produces globally optimal paths that A* cannot.</description><pubDate>Mon, 06 Apr 2026 00:00:00 GMT</pubDate><category>Pathfinding</category><category>GPU Compute</category><category>Eikonal</category><category>FIM*</category><category>wgpu</category><category>WGSL</category><category>A*</category><author>Anton Krupko</author></item><item><title>Confidence-Driven Sensor Fusion: When ML Disagrees with the Map</title><link>https://limina.agency/research/08-confidence-driven-sensor-fusion/</link><guid isPermaLink="true">https://limina.agency/research/08-confidence-driven-sensor-fusion/</guid><description>Maps show what was true last year. ML inference shows what might be true now. When they disagree, who wins? We built a GPU shader pipeline that resolves this question using confidence thresholds, hazard asymmetry, and entropy penalties.</description><pubDate>Mon, 06 Apr 2026 00:00:00 GMT</pubDate><category>Sensor Fusion</category><category>Machine Learning</category><category>SegFormer</category><category>GPU Compute</category><category>ONNX</category><category>Terrain Classification</category><author>Anton Krupko</author></item><item><title>CRDT Sync Over Constrained Radio: Fragmenting State Updates into 233-Byte Packets</title><link>https://limina.agency/research/09-crdt-sync-over-constrained-radio/</link><guid isPermaLink="true">https://limina.agency/research/09-crdt-sync-over-constrained-radio/</guid><description>How do you synchronize a collaborative CRDT document over LoRa radio with a 233-byte MTU and 1-5 kbps throughput? With custom fragmentation, BLAKE3 verification, and privacy-preserving routing — all designed for the worst network on Earth.</description><pubDate>Mon, 06 Apr 2026 00:00:00 GMT</pubDate><category>CRDT</category><category>Loro</category><category>LoRa</category><category>Meshtastic</category><category>Radio</category><category>Sync</category><category>Fragmentation</category><author>Anton Krupko</author></item><item><title>Designing for Gloves and Glare: Field UX for Emergency Services</title><link>https://limina.agency/research/10-designing-for-gloves-and-glare/</link><guid isPermaLink="true">https://limina.agency/research/10-designing-for-gloves-and-glare/</guid><description>When your user is strapped to a harness on a rooftop wearing gloves in strong glare, &apos;good UX&apos; means something different. How we reduced installation time from 7 minutes to under 3 — and NASA-TLX workload by 40% — by designing for exposure time, not tap count.</description><pubDate>Mon, 06 Apr 2026 00:00:00 GMT</pubDate><category>Field UX</category><category>IoT</category><category>Emergency Services</category><category>NASA-TLX</category><category>One-Handed Design</category><category>BLE</category><category>QR Codes</category><author>Anton Krupko</author></item><item><title>Keyboard-First Beats Mouse-First: Evidence from Drone Operations</title><link>https://limina.agency/research/11-keyboard-first-beats-mouse-first/</link><guid isPermaLink="true">https://limina.agency/research/11-keyboard-first-beats-mouse-first/</guid><description>A supervisory dashboard for hybrid sea-air drone logistics. Keyboard+mouse parity cut time-to-acknowledge by 50% and overall NASA-TLX workload by 47%. Here&apos;s the evidence and the design patterns.</description><pubDate>Mon, 06 Apr 2026 00:00:00 GMT</pubDate><category>Keyboard-First</category><category>Dashboard UX</category><category>NASA-TLX</category><category>Drone</category><category>Supervisory Systems</category><category>Accessibility</category><author>Anton Krupko</author></item><item><title>European Digital Sovereignty in Emergency Response: Why Architecture Matters More Than Flags</title><link>https://limina.agency/research/12-european-digital-sovereignty/</link><guid isPermaLink="true">https://limina.agency/research/12-european-digital-sovereignty/</guid><description>The argument for European digital sovereignty in critical infrastructure isn&apos;t nationalism — it&apos;s architecture. Zero-server, GDPR-native, open-protocol systems are structurally more sovereign than closed platforms regardless of where they&apos;re built.</description><pubDate>Mon, 06 Apr 2026 00:00:00 GMT</pubDate><category>European Sovereignty</category><category>GDPR</category><category>Defense Tech</category><category>ATAK</category><category>NATO</category><category>Critical Infrastructure</category><author>Anton Krupko</author></item><item><title>What Disaster Response Taught Us About Consumer Software</title><link>https://limina.agency/research/13-disaster-response-to-consumer-software/</link><guid isPermaLink="true">https://limina.agency/research/13-disaster-response-to-consumer-software/</guid><description>We built P2P mesh networking for mountain rescue. Then we used the same architecture to build a password manager and a notes app. The extreme constraints of disaster response produced infrastructure that&apos;s better for everyday use — here&apos;s why.</description><pubDate>Mon, 06 Apr 2026 00:00:00 GMT</pubDate><category>P2P</category><category>Kora</category><category>ARES</category><category>Architecture</category><category>Product Strategy</category><category>Consumer Software</category><author>Anton Krupko</author></item><item><title>GPU-Driven Terrain LOD: From Research to Production</title><link>https://limina.agency/blog/webgpu-terrain-lod/</link><guid isPermaLink="true">https://limina.agency/blog/webgpu-terrain-lod/</guid><description>How we moved terrain level-of-detail processing from the CPU to the GPU using WebGPU compute shaders, achieving a 21× speedup and enabling million-triangle terrain rendering in the browser.</description><pubDate>Thu, 15 Jan 2026 00:00:00 GMT</pubDate><category>WebGPU</category><category>3D Rendering</category><category>Performance</category><category>Compute Shaders</category><author>Limina Systems</author></item><item><title>ARES — Python Prototype (Legacy)</title><link>https://limina.agency/systems/ares-unified/legacy/</link><guid isPermaLink="true">https://limina.agency/systems/ares-unified/legacy/</guid><description>The original Project ARES: 5 months of research into disaster-zone pathfinding, the AI/physics pipeline, and the Python prototype that proved the concept — and revealed the five structural failures that demanded a Rust rewrite.</description><pubDate>Sun, 07 Dec 2025 00:00:00 GMT</pubDate><category>Product Design</category><category>Systems Thinking</category><category>Machine Learning</category><category>UX Research</category><category>Full-Stack</category><category>GEOINT</category><category>Pathfinding</category><category>Real-Time Systems</category><author>Limina Systems</author></item><item><title>Building ARES: Lessons from Geospatial AI for Humanitarian Routing</title><link>https://limina.agency/blog/building-ares-geoint/</link><guid isPermaLink="true">https://limina.agency/blog/building-ares-geoint/</guid><description>How we designed a geospatial intelligence platform that combines satellite imagery, drone data, and physics-based AI to guide humanitarian convoys through post-disaster terrain.</description><pubDate>Fri, 05 Dec 2025 00:00:00 GMT</pubDate><category>Geospatial</category><category>AI</category><category>Humanitarian</category><category>Systems Engineering</category><author>Limina Systems</author></item><item><title>Real-Time Terrain Mesh LOD via Compute Shader Decimation</title><link>https://limina.agency/research/terrain-lod/</link><guid isPermaLink="true">https://limina.agency/research/terrain-lod/</guid><description>GPU-driven continuous level-of-detail system for large-scale terrain visualization. Compute shader evaluates screen-space error per patch and emits variable-density index buffers in a single dispatch.</description><pubDate>Mon, 20 Oct 2025 00:00:00 GMT</pubDate><category>WebGPU</category><category>Compute Shaders</category><category>3D Rendering</category><category>LOD</category><author>Limina Systems</author></item><item><title>Usage-First Systems Architecture for Smart City Sensors</title><link>https://limina.agency/systems/sensor-install/</link><guid isPermaLink="true">https://limina.agency/systems/sensor-install/</guid><description>Mobile app and system design enabling quick field sensor installation and server pairing via QR codes and Bluetooth Low Energy (BLE).</description><pubDate>Fri, 10 Oct 2025 00:00:00 GMT</pubDate><category>Product Design</category><category>Mobile App</category><category>IoT</category><category>BLE</category><category>QR Codes</category><category>Onboarding</category><category>UX</category><category>CAD</category><category>System Design</category><author>Limina Systems</author></item><item><title>Why Offline-First Matters for Field Deployments</title><link>https://limina.agency/blog/offline-first-field-apps/</link><guid isPermaLink="true">https://limina.agency/blog/offline-first-field-apps/</guid><description>Connectivity assumptions that work in urban environments fail catastrophically in field deployments. Here&apos;s why offline-first architecture isn&apos;t optional for sensor networks — and how CRDTs make it practical.</description><pubDate>Sat, 20 Sep 2025 00:00:00 GMT</pubDate><category>Offline-First</category><category>IoT</category><category>Distributed Systems</category><category>CRDT</category><author>Limina Systems</author></item><item><title>Adaptive RF Denoising for Low-SNR Environments</title><link>https://limina.agency/research/rf-denoising/</link><guid isPermaLink="true">https://limina.agency/research/rf-denoising/</guid><description>Wavelet-domain neural denoising pipeline for HF/VHF radio receivers operating below noise floor. Achieves 8dB effective SNR gain on COTS SDR hardware through learned spectral priors.</description><pubDate>Fri, 15 Aug 2025 00:00:00 GMT</pubDate><category>Signal Processing</category><category>Machine Learning</category><category>Radio Systems</category><category>Edge Computing</category><author>Limina Systems</author></item><item><title>Drone Monitoring Dashboard</title><link>https://limina.agency/systems/drone-monitoring-dashboard/</link><guid isPermaLink="true">https://limina.agency/systems/drone-monitoring-dashboard/</guid><description>Keyboard-first, map-centric supervisory dashboard for a hybrid sea-air (UAV + maritime) delivery network in Hong Kong.</description><pubDate>Thu, 31 Jul 2025 00:00:00 GMT</pubDate><category>Systems UX</category><category>IoT/Robotics</category><category>Maritime</category><category>UAV</category><category>Keyboard-First</category><category>Accessibility</category><author>Limina Systems</author></item><item><title>Conflict-Free Sync for Offline-First Field Applications</title><link>https://limina.agency/research/offline-sync/</link><guid isPermaLink="true">https://limina.agency/research/offline-sync/</guid><description>CRDT-based synchronization protocol for field-deployed sensor networks with intermittent connectivity. Guarantees eventual consistency with sub-second merge on reconnect across 500+ node clusters.</description><pubDate>Tue, 10 Jun 2025 00:00:00 GMT</pubDate><category>CRDT</category><category>Distributed Systems</category><category>Offline-First</category><category>IoT</category><author>Limina Systems</author></item><item><title>Applications</title><link>https://limina.agency/systems/kora/apps/</link><guid isPermaLink="true">https://limina.agency/systems/kora/apps/</guid><description>Kora Vault (P2P password manager) and Kora Notes (collaborative editing) — two apps that prove the framework.</description><pubDate>Sun, 01 Jun 2025 00:00:00 GMT</pubDate><category>Rust</category><category>Tauri</category><category>P2P</category><category>Password Manager</category><category>Collaborative Editing</category><category>CRDT</category><category>E2E Encryption</category><author>Limina Systems</author></item><item><title>Engineering Reference</title><link>https://limina.agency/systems/kora/technical/</link><guid isPermaLink="true">https://limina.agency/systems/kora/technical/</guid><description>8-crate Rust workspace: Iroh QUIC transport, vector-clock CRDT sync, Ed25519 identity mesh, E2E encryption pipeline.</description><pubDate>Sun, 01 Jun 2025 00:00:00 GMT</pubDate><category>Rust</category><category>Iroh</category><category>Loro</category><category>CRDT</category><category>QUIC</category><category>Ed25519</category><category>ChaCha20</category><category>WASM</category><category>P2P</category><author>Limina Systems</author></item><item><title>Vision</title><link>https://limina.agency/systems/kora/vision/</link><guid isPermaLink="true">https://limina.agency/systems/kora/vision/</guid><description>Why the cloud custody model is broken, how P2P can fix it, and the design decisions behind Kora.</description><pubDate>Sun, 01 Jun 2025 00:00:00 GMT</pubDate><category>Rust</category><category>P2P</category><category>Iroh</category><category>CRDT</category><category>E2E Encryption</category><category>Systems Design</category><category>Privacy-First</category><author>Limina Systems</author></item><item><title>ARES Unified — Story</title><link>https://limina.agency/systems/ares-unified/journey/</link><guid isPermaLink="true">https://limina.agency/systems/ares-unified/journey/</guid><description>From GPS failure in disaster zones to a production P2P mesh C2 platform: the full story of ARES — why it was built, what the prototype discovered, what it could not do, and how the 18-crate Rust rewrite resolved every structural failure.</description><pubDate>Sat, 01 Feb 2025 00:00:00 GMT</pubDate><category>Rust</category><category>Tauri</category><category>P2P Mesh</category><category>Iroh</category><category>Loro</category><category>Systems Design</category><category>Offline-First</category><category>Product Design</category><author>Limina Systems</author></item><item><title>ARES Unified — Engineering Reference</title><link>https://limina.agency/systems/ares-unified/technical/</link><guid isPermaLink="true">https://limina.agency/systems/ares-unified/technical/</guid><description>Complete technical documentation: 18-crate Rust workspace, Iroh P2P mesh, Loro CRDT sync, TVPE voxel encoding, FIM* GPU pathfinding, OSM+ML material fusion, radio transport, ATAK interop, and every implementation detail.</description><pubDate>Sat, 01 Feb 2025 00:00:00 GMT</pubDate><category>Rust</category><category>Iroh</category><category>Loro</category><category>wgpu</category><category>CRDT</category><category>P2P</category><category>FIM*</category><category>SSVG</category><category>ONNX</category><category>Tauri</category><category>MAVLink</category><category>Meshtastic</category><author>Limina Systems</author></item></channel></rss>