THE A818 ECOSYSTEM
The A818 designation encompasses a suite of proprietary projects spanning neurotechnology, electromagnetics, and legacy software governance.
A818_SK (Sky Net) — Electromagnetic Effects & Threat Analysis
A high-complexity, real-time drone detection and EMP effects analysis software platform. Developed from first principles, this system integrates computer vision, geospatial tracking, and published EMP physics formulas into a unified desktop application.
Core Architecture: Built in Python 3.13, utilizing a Tkinter GUI (1350x720), OpenCV for live webcam feeds, and Matplotlib for real-time graphing.
Computer Vision: Implements YOLOv8 for target detection and coordinate extraction, currently operating on CPU inference (Intel Core i5-5200U). Training pipelines are configured for Kaggle/Colab T4 environments.
EMP Physics Engine: A custom-built physics module (a818em) that calculates electromagnetic effects using standardized HEMP geometry. It processes coverage radius, geometric spreading attenuation, Poynting vector energy density, and chip-coupled energy thresholds.
Geospatial & Location: Features a database of 150,513 world cities and utilizes IP-API for real-time location detection, calculating geographical distances via the Haversine formula.
Hardware Abstraction: Includes a localized hardware interface (hardware.py) using pyserial for line-based ASCII serial control. It features a robust state machine (IDLE → CHARGING → READY → ARMED → FIRING → COOLDOWN) with built-in safety interlocks and auto-disarm protocols.
Current Status (60% Complete): The detection pipeline, GUI, GPS mapping, live graphing, and EMP physics engines are fully operational and verified. The system is currently processing a 9.14 GB drone-specific dataset (Seraphim) to train the YOLOv8 model from COCO (80 classes) to drone-specific (1 class), targeting an mAP50 of ≥ 0.98. Hardware control scripts are written and secure but remain untested pending physical device availability.
A818_NF (NeuroFlow) — Neurotechnology & Signal Processing
A software-first Brain-Computer Interface system designed to decode motor intent from raw electroencephalogram (EEG) signals. This system was engineered independently, utilizing public datasets and custom-built machine learning pipelines.
Architecture: Utilizes a custom implementation of the EEGNet architecture and Common Spatial Patterns (CSP) for feature extraction.
Performance: Achieves 91% accuracy on 3-class intent decoding, and 78% accuracy on 5-class decoding with per-subject calibration.
Status: The software infrastructure is complete, fully open-source, and deployed live. Hardware integration (an EEG headband) is currently in active development, with components being sourced and integrated from local markets. This ensures a ground-up understanding of the physical device interface abstraction.
Technical Architecture & Pipeline (A818_NF)
Focus Mode (3 Classes): Rest, Left Fist, Right Fist. Accuracy: 91%. Uses Logistic Regression. No calibration required.
Extended Mode (5 Classes): Adds Both Fists, Both Feet. Accuracy: 78%. Uses EEGNet CNN. Requires 20-minute per-subject calibration.
Data Processing Pipeline: Raw EEG (64 channels × 481 samples) → 50 Hz notch filter → 8–30 Hz bandpass → per-channel z-score normalization → classifier → intent prediction.
Engineering Principles: The Live Recording mode refuses to run without a physical headband connected. No fake predictions. Bench Test mode shows ground truth on every prediction (correct or incorrect).
Stack: Python, PyTorch, scikit-learn, MNE-Python, Streamlit.
Dataset: PhysioNet EEG Motor Movement/Imagery (109 subjects).
A818_K — Legacy Code Governance
The next scheduled deployment within the ecosystem. A818_K is a legacy code mapping and governance tool designed to analyze, map, and manage complex, inherited software architectures. Development begins immediately upon the shipping of A818_SK.