makalin

Stage	Key Actions	Technologies & Techniques	Output
1. Data Collection & Preprocessing	Gather data from web, internal docs, APIs. Clean, deduplicate, and standardize formats.	Web scraping, API calls, data clea

makalin

Feature	VidNCI (Active Detection)	Passive Detection Libraries (e.g., FaceForensics++)
Core Methodology	Proactively embeds a verifiable signal (noise-coded illumination) at the time of video capture.	Reactively analyzes a video after creation, searching for artifacts left by the generative process.
Detection Basis	Correlation with a known, pre-defined pseudo-random code.	Classification based on learned features and artifacts from training data.
Robustness to Compression	High. The signal is designed to be robust against common post-processing.	Low to Medium. Performance often degrades with compression like H.264.
Vulnerability to Adversarial Attacks	Low. Requires physical replication of the coded light, a much harder task.	High. Attackers can train models to minimize the very artifacts detectors look for.
Generalizability	High. The method is independent of the specific deepfake generation technique.	Low. Models often fail o

makalin

Step	Name	Description	Key Action
1	Code Generation	A unique, pseudo-random noise code is generated. This code serves as the invisible signature for the video.	Creating a cryptographically secure pattern (e.g., Gaussian or binary noise) that will be used to modulate the light source.
2	Video Embedding	The generated code is used to modulate the brightness of a light source in the scene. This coded light illuminates the environment during recording.	Physically embedding the signature into the light that is captured by the camera, making it an intrinsic part of the video's pixel data.
3	Code Extraction	The VidNCI library analyzes the recorded video to isolate and recover the embedded noise code from the pixel data.	Using signal processing techniques to extract the faint signature from the video frames, resulting in a "code image."
4	Analysis	The extracted code is compared against the original, known code. A

makalin

Feature	FRESCO	AVIF	WebP	JPEG XL
Lossy Compression	State-of-the-art	Excellent	Good	Very Good
Lossless Compression	State-of-the-art	Good	Good	Excellent
Transparency	Yes	Yes	Yes	Yes
Animation	Yes (efficient)	Yes	Yes	Yes
HDR/WCG Support	Yes	Yes	No	Yes
3D Model Support	Yes (native)	No	No	No
Vector Graphics	Yes (native)	No	No	No
Encoding Speed	Optimized	Slow	Fast	Moderate

makalin

Threat Vector	Concrete Risk
OAuth Token Leakage	LLM logs or tool responses accidentally dump a valid JWT with write scopes.
Prompt Injection	A malicious prompt convinces the LLM to run delete_everything().
Rogue MCP Server	A cloned server with the same name but evil tools.
Tool Poisoning	A benign tool is replaced with a back-doored version.

makalin

Technology	Input Method	Output	Personalization	Examples
Sonit	Non-verbal vocal gestures	Text or actions	High (machine learning-based)	Murmurs, hums, cultural sounds
Proloquo2Go	Symbol selection	Synthesized speech	Moderate (customizable symbols)	iPad-based app
PECS	Picture card exchange	Physical or verbal response	Low (pre-defined cards)	Picture Exchange Communication System
SGDs (e.g., Lingraphica)	Text or symbol input	Synthesized speech	Moderate (customizable phrases)	Tablet-based devices
Motion Sensing	Gestures/body movements	Digital commands	Moderate (pre-trained models)	Kinect-based systems

makalin

Component	Technology	Purpose
Core Logic	Python	Handles application logic and training pipeline
GUI	Kivy	Provides a lightweight, cross-platform interface for mobile and desktop
Machine Learning	PyTorch	Powers deep learning models for sound classification
Audio Processing	NumPy, Librosa	Processes audio signals for analysis
Data Storage	SQLite	Stores user-specific training data locally

makalin

🔧 Application	🌟 Neuromorphic Edge	🧪 Example
Robotics	Real-time reflexes	Intel Loihi
Edge AI / IoT	Extreme power savings	BrainChip Akida
Self-Driving Cars	Fast reaction times	IBM NorthPole
Medical Imaging	Low-power diagnosis	TrueNorth
Telecom Optimization	Efficient AI models	Loihi 2 at Ericsson

makalin

🧠 Technology	📌 Key Idea	⚡ Superpower
Spiking Neural Networks	Brain-like event-driven processing	Ultra-low power, fast on temporal data
Memristors	Adaptive synapses with memory	Real-time learning, persistent memory
In-Memory Computing	Merge of storage and processing	No data shuffling, ultra-low latency

makalin

function sanitizeInput($input) {
    return htmlspecialchars(strip_tags(trim($input)), ENT_QUOTES, 'UTF-8');
}

// Usage:
$userInput = "<script>alert('hack');</script>";
$safeInput = sanitizeInput($userInput);
echo $safeInput; // Output will be sanitized