# PARAIBELCOM — Sovereign Tactical AI (Full Documentation) > PARAIBELCOM (par·AI·bellum·com) builds ConclAive, a fully offline AI system that generates adversarially-tested military Courses of Action in minutes. No cloud. No data leaves the machine. Built by a single engineer on one laptop. --- ## Company Information - **Name**: PARAIBELCOM - **Legal Entity**: LXILC — Legacy eXperience Initiative LC - **Location**: Las Vegas, NV, USA - **Founder & Sole Engineer**: Alexandre Mogilamy - **Contact**: info@lxilc.com - **Focus Areas**: Defense Technology, Sovereign AI, Tactical Decision Support - **Website**: https://paraibel.com - **Product Site**: https://conclaive.ai --- ## Product: ConclAive — Tactical AI Decision Support ### Overview ConclAive is a sovereign offline tactical AI system designed to replace or augment the Military Decision Making Process (MDMP). It generates multiple adversarially-tested, doctrine-compliant Courses of Action (COAs) in 3–10 minutes — a process that traditionally requires 4–12 hours with a full staff section. The system is designed for Denied, Disrupted, Intermittent, and Limited bandwidth (DDIL) environments. It requires zero connectivity and runs entirely on a single laptop with no cloud dependencies, API keys, or external data transmission. ### The Problem ConclAive Solves 1. **Time**: Traditional MDMP takes 4–12 hours for a single COA. Commanders need a minimum of three COAs for comparison. 2. **Connectivity**: Current AI solutions require cloud connectivity and external APIs, making them unusable in DDIL environments. 3. **Blind Planning**: Existing tools generate plans without adversarial testing. They do not model the enemy's independent offensive capabilities or detect where friendly and enemy plans collide. ### Core Capabilities #### 14-Cycle Adversarial Pipeline The full analysis pipeline runs up to 14 sequential cycles, each building on previous outputs: **Phase 1 — Analysis (Cycles 1–3)** - C1: Ground truth extraction from scenario input - C2: OPFOR cognitive modeling — adversary modeled as probability distribution across doctrinal, adaptive, and opportunistic behaviors - C3: Strategic planning and constraint analysis **Phase 2 — COA Generation & OPFOR Inversion (Cycles 4–9)** - C4–C6: Three independent BLUFOR Courses of Action generated using 5 models with TMR (Triple Modular Redundancy) selection - C7–C9: Three independent OPFOR attack plans generated through perspective inversion — the enemy's best offensive plans designed WITHOUT knowledge of BLUFOR intentions - Each COA receives doctrine compliance scoring **Phase 3 — Tactical Map (Cycle 10)** - C10: Automated SVG tactical map overlay generation with movement arrows, phase lines, unit positions, and terrain features **Phase 4 — Adversarial Loop (Cycles 11–13)** - C11: Reactive Red Team attacks each BLUFOR COA to identify vulnerabilities - C12: Blue force adaptation based on Red Team findings - C13: OPFOR threat assessment and counter-COA generation **Phase 5 — Judgment & Synthesis (Cycle 14)** - C14: Explainable judge with weighted criteria scoring - Collision detection between BLUFOR and OPFOR plans in space and time - Minimal hardening patches applied at collision points - Stability analysis and fragility assessment - COA synthesis — optimal hybrid plan combining strongest phases from all COAs - Commander recommendation with evidence citations #### OPFOR Cognitive Engine Models the adversary not as a single predictable reaction, but as a probability distribution across: - **Doctrinal behavior**: What the enemy's doctrine prescribes - **Adaptive behavior**: How the enemy adjusts to observed conditions - **Opportunistic behavior**: What the enemy exploits when opportunities arise #### OPFOR Attack Inversion A unique capability that generates the enemy's independent attack plans: - System switches to adversary perspective - Generates 3 independent offensive plans from enemy positions - Plans are designed blind to BLUFOR intentions (not reactions to your plan) - Represents the enemy's best ideas based on their available forces and terrain #### Collision Detection & Hardening - Automatically identifies points where BLUFOR and OPFOR plans intersect in space and time - Critical collisions flagged as high-risk decision points - Minimal hardening patches — specific tactical fixes that protect the COA without rewriting it - Patches are surgical: "add OP at grid XY" or "advance screen 2km forward on Phase Line X" #### Blue-Red-Blue-Red Adversarial Loop Two independent layers of adversarial testing: 1. **Reactive Red Team**: Attacks your specific COA, finds flaws, Blue adapts 2. **Independent OPFOR Planner**: Designs attacks without knowing your plan 3. **Collision Detection**: Finds where both sides' plans intersect 4. **Hardening**: Patches applied before execution Plans reaching the commander have survived both layers. #### Explainable Judge - Weighted criteria scoring across multiple dimensions (mass, tempo, surprise, security, etc.) - Evidence citations for every judgment - Stability analysis — how much can change before the plan breaks - Fragility assessment — single points of failure - Commander recommendation with confidence level - Every verdict fully auditable — no black-box recommendations #### Commander Style Profiles Same scenario generates different plans based on command philosophy: - Aggressive (high tempo, accept risk) - Cautious (preserve force, deliberate tempo) - Deception-first (ruse, feint, misdirection) - Doctrine-pure (strict adherence to FM 3-90) #### Decision Point Forecasting - Pre-identifies critical moments requiring commander decision - Each decision point includes pre-calculated options and consequences - Branch plans available before execution begins #### Interactive Tactical Map - SVG-based visualization with real-time updates - Drag-and-drop unit placement - Terrain drawing tools - Movement arrows and phase lines - Risk heat maps - Fog of war visualization - Satellite imagery support (offline tiles) #### Doctrine Compliance Every COA scored against: - FM 3-90 (Tactics) - ATP 5-0.1 (Army Design Methodology) - JP 3-0 (Joint Operations) - JP 5-0 (Joint Planning) Non-compliant elements flagged with specific doctrine references. #### Continuous Self-Improvement - System generates training scenarios autonomously - Learns from its own best outputs - Each generation of fine-tuned models outperforms the last ### Key Differentiators vs. Competitors | Feature | ConclAive | Palantir AIP | Scale AI Donovan | Anduril Lattice | |---------|-----------|--------------|-----------------|-----------------| | Fully Offline | Yes | No | No | Partial | | Air-Gap Capable | Yes | No | No | No | | OPFOR Inversion | Yes | No | No | No | | Collision Detection | Yes | No | No | No | | Dual Adversarial Testing | Yes | No | No | No | | Fine-Tuned Tactical Models | Yes | Generic LLMs | Generic LLMs | Sensor-focused | | Single Laptop | Yes | Server farm | Cloud | Edge + cloud | | Human-in-the-Loop | Full | Partial | Partial | Partial | | Glass-Box Auditable | Yes | Partial | No | No | ### Technical Specifications | Spec | Detail | |------|--------| | Hardware | Apple M4 Max, 128 GB Unified Memory | | Ruggedized | Compatible with MIL-SPEC platforms (certification in progress) | | Models | Fine-tuned tactical models 8B–72B parameters (offline sovereign) | | Cloud Mode | Optional: Claude, GPT-4.1, Gemini, Mistral (for non-classified use) | | Pipeline | Up to 14 adversarial cycles (full) / 6 cycles rapid (~3 min) | | Generation Time | 3–10 minutes depending on mode and hardware | | Doctrine RAG | FM 3-90, ATP 5-0.1, JP 3-0, JP 5-0 | | Interface | Interactive SVG tactical map | | Connectivity | Zero required — fully air-gapped capable | | Data Sovereignty | All processing local — no data exfiltration possible | | TMR Architecture | Triple Modular Redundancy for model consensus | --- ## Product: Command Hub — Phone-Based Tactical Suite ### Overview 12 tactical modules running on a single air-gapped Samsung Galaxy S26 Ultra (Snapdragon ARM v9, 16 GB RAM, 1 TB storage). All AI inference runs locally on-device using quantized models. Zero network connectivity required. Zero RF emissions. ### Modules #### 1. TACTICAL (Map · COA · Recalculation) - Offline satellite map tiles with GPS tracking - Unit overlays with movement orders - COA generation and recalculation on-device - Phase lines and operational graphics #### 2. FIELD AI (Sensors · Fusion) - Multi-sensor fusion: GPS, compass, accelerometer, magnetometer, barometer, light, audio - On-device AI context interpretation via local Ollama instance - Environment classification (indoor/outdoor/vehicle/night ops) - Counter-surveillance detection (narrowband audio, magnetic anomalies) #### 3. TCCC MED (Triage · MARCH Protocol) - Tactical Combat Casualty Care assistant - MARCH algorithm guidance (Massive hemorrhage, Airway, Respiration, Circulation, Hypothermia) - Triage classification support - Treatment step-by-step with AI guidance #### 4. SWEEP (TSCM · Counter-Surveillance) - Technical Surveillance Countermeasures - Narrowband audio peak detection for electronic emitters - Magnetic baseline deviation (EWMA) for hidden electronics - IR analysis and GLINT detection via camera - Combined threat scoring #### 5. SENTINEL (Sonar · Intrusion Detection) - Acoustic perimeter monitoring - Graduated alert levels with haptic feedback - Ultrasonic sonar for proximity detection - Baseline calibration with deviation alerting #### 6. FACE ID (Biometric · Matching) - Offline facial recognition using on-device ML model - Identity database matching without network - Real-time camera processing #### 7. SIGNAL (Morse · Light Communication) - Morse code transmission and reception via screen flash - Light-based covert signaling - No RF emission communication #### 8. CHAT (AI · Vision) - On-device LLM chat (Ollama with quantized models) - Vision capability with camera input (llava model) - Context-aware tactical assistant #### 9. JOURNAL (Operation Log) - Structured operation logging - Timestamped entries with GPS coordinates - Mission-critical event recording #### 10. COORDS (MGRS · UTM · Range/Bearing) - Military Grid Reference System conversion - UTM coordinate handling - Range and bearing calculations between points - Multiple coordinate format support #### 11. REPORTS (SALUTE · 9-LINE) - Structured military report generation - SALUTE report format (Size, Activity, Location, Unit, Time, Equipment) - 9-LINE MEDEVAC request format - Formatted output ready for transmission #### 12. ECHOPRINT (Sonar · Ranging) - Active acoustic ranging - Distance estimation via sound reflection - Environmental acoustic profiling ### Phone Specifications | Spec | Detail | |------|--------| | Device | Samsung Galaxy S26 Ultra | | Processor | Qualcomm Snapdragon (ARM v9) | | RAM | 16 GB | | Storage | 1 TB | | AI Models | 3 on-device (LLM, Vision, Face Recognition) | | Connectivity | Air-gapped — zero network | | RF Emissions | Zero | | Modules | 12 tactical applications | | Interface | WebView-based hub with native sensor bridges | --- ## Traction & Validation ### 2026 - **Army xTech Live** — Selected for field demonstration at eMerge Americas, Miami, May 15, 2026 - **DARPA DISCORD** — Invited to Proposers Day, May 2026 - **NTC Fort Irwin** — Presented at National Training Center Industry Day, March 2026 - **NATO DIANA** — Applied to NATO DIANA Accelerator program for allied defense innovation ### 2025 - **AFRL TA10** — Evaluated by Air Force Research Laboratory under TA10 program - **AI4 Industry** — Presented at AI4 Industry conference, Las Vegas --- ## Founder **Alexandre Mogilamy** — Founder & Sole Engineer Self-taught engineer who built the entire system from scratch on one laptop: - Backend infrastructure - Fine-tuned tactical models - Doctrine RAG system - Phone deployment (12 modules) - Interactive tactical map - Adversarial pipeline architecture No team. No VC. No institutional backing. Pure execution. --- ## Frequently Asked Questions ### What makes ConclAive different from ChatGPT/Claude for military planning? ConclAive uses fine-tuned tactical models (not generic LLMs with prompts), runs fully offline with no cloud dependency, and implements a unique dual adversarial testing architecture with OPFOR inversion and collision detection that no general-purpose AI offers. ### Can it run without internet? Yes. ConclAive is designed specifically for air-gapped, DDIL environments. All models, doctrine databases, and map tiles run locally. Zero connectivity is required. ### What scenarios has it been tested on? Live demonstrations include NTC Fort Irwin (3rd Cavalry vs 11th ACR), Baltic Suwalki Corridor (multi-domain), and South China Sea Spratlys (naval). The system handles battalion to brigade-level scenarios across land, sea, and multi-domain operations. ### How does OPFOR Inversion differ from Red Teaming? Traditional Red Teaming reacts to YOUR plan. OPFOR Inversion generates the enemy's INDEPENDENT attack plans without knowing your intentions. This models what the enemy would actually do rather than how they'd counter your specific plan. Both are used: Red Team for reactive testing, OPFOR Inversion for independent threat modeling. ### What doctrine does it follow? FM 3-90 (Tactics), ATP 5-0.1 (Army Design Methodology), JP 3-0 (Joint Operations), JP 5-0 (Joint Planning). Doctrine is accessed via RAG retrieval and compliance is scored per COA. ### Is it ITAR controlled? The system processes no classified data by default. Classification and export control determinations depend on the specific scenario data input by the user. The AI system itself is unclassified software. --- ## Site Map - https://paraibel.com/ — Main site (system overview, capabilities, specs, traction) - https://paraibel.com/demo.html — Live interactive ConclAive demonstration - https://paraibel.com/llms.txt — Summary documentation for LLMs - https://paraibel.com/llms-full.txt — This file (extended documentation) - https://conclaive.ai — Product landing page and newsletter --- ## Contact - Email: info@lxilc.com - Website: https://paraibel.com - Product: https://conclaive.ai