Apppublic

Antares-Environments/WeltVX

sourceHugging Faceupdated 7mo agoView on Hugging Face

0likes

App README

Welt VX 3.1

Redefining Viewer Experience with Relative Video Intelligence. Developed by Antares Agentic Environments

Welt VX 3.1 is an event-driven, multimodal video intelligence application. It transforms passive video playback into an interactive, agent-assisted workspace. Built on the modern Google GenAI SDK and the Reflex framework, the system allows users to cross-analyze multiple video files, generate smart chapters, transcribe audio, and instantly jump to semantic timestamps using natural language.

Architecture: The V-Map

Welt VX operates on a strictly defined, three-tier architecture ensuring resilience, state isolation, and deterministic UI updates.

1.stability.py (The Infrastructure): Manages network resilience via exponential backoff matrices and enforces strict JSON validation firewalls.
2.engine.py (The Logic Core): Houses the intelligence engine powered by Gemini 3 Flash Preview, executing contextual analysis and routing.
3.state.py (The Brain): Manages the Reflex user interface, multi-tenant UUID session isolation, background event locking, and dynamic asset routing.

The 1+7 Task Force

The intelligence engine operates via an Orchestrator and seven specialized agents, executing deterministic functions based on user intent.

Agent	Function	Output
Orchestrator	Intent Router	Analyzes user queries and routes them to the correct specialist.
The Scribe	Audio Transcription	Generates perfectly timed HTML-styled SRT subtitle files.
The Librarian	Asset Management	Analyzes relative video context and auto-labels multiple files.
The Architect	Narrative Flow	Generates timeline-segmented Smart Chapters.
The Detective	Forensic Scanning	Identifies specific visual/audio instances with exact timestamps.
The Navigator	Semantic Seeking	Triggers the video player to jump directly to a requested event.
The Guide	Context Analyst	Answers questions based purely on the provided media context.
The Mechanic	Data Repair	Implements Human-in-the-Loop fixes for subtitles or JSON schemas.

Key Features

—Relative Video Intelligence: Upload multiple video assets simultaneously to compare, contrast, and establish chronological connections across different files.
—Zero-Latency Native Subtitles: Generates plain-text SRT files that the browser natively mounts over the video player without requiring heavy synchronization logic.
—Firestore Feedback Loop: Logs user-corrected outputs from The Mechanic to a Google Cloud database to continuously refine system accuracy.

Deployment & Usage

1. Cloud Infrastructure

Welt VX is optimized for Dockerized deployment on platforms providing substantial RAM, such as huggingface.co.

Required Environment Variables:

—SHARED_API_KEY: Your Gemini API Key.
—VX_MASTER_PASSWORD: Application access gate.
—GCP_CREDENTIALS_JSON: Raw JSON string of your Google Cloud Service Account (for Firestore feedback logging).

2. Local Initialization

To compile and run the application locally on your machine:

1.Clone the repository.
2.Install dependencies: pip install -r requirements.txt
3.Initialize the Reflex environment: reflex init
4.Launch the application: reflex run

(Note: Ensure your local environment variables are configured securely prior to launching).

License

See the LICENSE file in the repository root for usage terms and distribution rights.