Video RSS Aggregator

Intelligent video summarization and RSS feed generation powered by Qwen3 models on NVIDIA CUDA.

• ASR: Qwen/Qwen3-ASR-1.7B (via qwen-asr)
• Summarization: Qwen/Qwen3-8B-AWQ (via vLLM)
• Storage: PostgreSQL
• API: FastAPI

Requirements

• Python 3.11+
• NVIDIA GPU with CUDA (Windows / Linux)
• PostgreSQL 15+
• ffmpeg on PATH

Quick Start

# Create environment
python -m venv .venv
.venv\Scripts\activate   # Windows
# source .venv/bin/activate  # Linux

# Install
pip install -e .

# Configure
set DATABASE_URL=postgresql://user:pass@localhost:5432/video_rss

# Run
python -m vra serve --bind 0.0.0.0:8080

Models are downloaded from Hugging Face automatically on first run.

Environment Variables

Variable	Default	Description
DATABASE_URL	(required)	PostgreSQL connection string
BIND_ADDRESS	0.0.0.0:8080	HTTP server bind address
API_KEY	(none)	Optional bearer token for auth
VRA_STORAGE_DIR	.data	Local storage for downloads and audio
VRA_ASR_MODEL	Qwen/Qwen3-ASR-1.7B	ASR model name or path
VRA_LLM_MODEL	Qwen/Qwen3-8B-AWQ	Summarization model name or path
VRA_GPU_MEMORY_UTILIZATION	0.8	vLLM GPU memory fraction
VRA_ASR_DEVICE	cuda:0	PyTorch device for ASR
VRA_ASR_MAX_TOKENS	4096	Max tokens for ASR output
VRA_LLM_MAX_TOKENS	2048	Max tokens for summarization
VRA_RSS_TITLE	Video RSS Aggregator	RSS feed title
VRA_RSS_LINK	http://localhost:8080/rss	RSS feed self-link
VRA_RSS_DESCRIPTION	Video summaries	RSS feed description

API

GET /health

Returns health status.

POST /ingest

Ingest an RSS/Atom feed. Body: {"feed_url": "...", "process": true, "max_items": 5}

POST /process

Process a single video/audio source. Body: {"source_url": "...", "title": "..."}

GET /rss?limit=20

Returns summarized content as RSS 2.0 XML.

Verification

python -m vra verify --feed-url "https://example.com/feed.xml" --source "/path/to/audio.wav"