Digital Liveness & Forensics Suite (WIP)

Status: Actively Under Development. > The isolated Python forensic microservices are currently being built and tested. The central Java Spring Boot API Gateway and JavaScript Client-Side Tracker are pending integration.

Main Purpose

A polyglot, microservice-based fusion engine designed to verify human interaction and detect AI-generated media (Deepfakes, LLM text, cloned audio, and synthesized images).

Architecture Overview

This system utilizes a Scatter-Gather pattern. A central Java Gateway routes media to isolated, domain-specific Python forensic services, fusing their confidence scores into a final "Probability of Human Origin."

Completed Microservices (Isolated Testing Phase)

• Biological Liveness Service (Python/FastAPI): Uses rPPG and MediaPipe to detect micro-fluctuations in facial blood volume.
• Audio Forensics Service (Python/FastAPI): Uses Wav2Vec2 and librosa to detect missing breath sounds, phase irregularities, and unnatural MFCC acceleration in synthetic voices.
• Reverse Engineering Service (Python/FastAPI): Analyzes raw hex streams in chunks to prevent Zip Bombs while detecting programmatic compiler fingerprints (e.g., automated FFmpeg wrappers).
• Vision Artifact Service (Python/FastAPI): Uses Vision Transformers (ViT) and FFT analysis with adversarial blurring to detect latent diffusion tiling.
• Text & NLP Service (Python/FastAPI): Uses local GPT-2 deterministic math to measure sentence perplexity and burstiness variance to catch LLM-generated text.

Pending Microservices

• API Gateway / Central Brain (Java Spring Boot): The orchestrator utilizing Resilience4j circuit breakers and async routing.
• Client-Side Behavioral Tracker (JavaScript): Detects botnet keystroke dynamics and mouse trajectories.
• Metadata & Provenance Service (Java): Cryptographic C2PA and EXIF header inspection.

Prerequisites & Setup

Because this suite relies heavily on system-level signal processing and local AI inference, standard package managers are not enough.

1. System Dependencies: You must install ffmpeg on your host machine (via winget, brew, or apt) to decode complex audio/video codecs.
2. Local AI Models: To prevent runtime crashes and save bandwidth, the Hugging Face models (Wav2Vec2, ViT, GPT-2) must be downloaded manually into their respective local_*_model directories before starting the FastAPI servers.
3. Docker: Required for the Reverse Engineering service to enforce a strict 512MB RAM quota against decompression attacks.

Author

Niroshan Dh