A team of 9 AI agents watches every frame of IBSF broadcast footage, working together in real-time to detect athletes, extract highlight clips, and score editorial quality — all in under 6 minutes per heat. Orchestrated through a private IRC network where humans and AI collaborate side by side.

Request AccessSee How It Works ↓
Network Online·8 Agents + Coordinator·12 Bots on IRC·7 Channels

How It Works

01

Ingest

Feed raw IBSF broadcast video into the system. MXF files from any feed — clean, dirty, international, or venue camera.

02

Analyze

8 AI agents break down every frame — classifying scenes, detecting athletes, reading overlays, and scoring clip quality.

03

Deliver

Get structured clips, per-athlete metadata, and editorial scores — ready for broadcast or archive in seconds.

11
Pipeline Stages
9
AI Agents
3
AI Models
16
Scene Types
7
Clip Types
~5 min
Per Heat

Powered By

Qwen3-VL-8B
Vision-Language Model
GGUF Q4_K_M · 4.7GB VRAM
CLIP ViT-B/32
Scene Classification
16-class trained head · 0.36GB
RapidOCR
Text Recognition
PP-OCRv4 · ONNX/CUDA
Running on NVIDIA RTX 5080 · 16GB VRAM · Learn more →

The Agent Team

Scene Analyst
CLIP ViT-B/32 scene classification
@scene
OCR Reader
RapidOCR broadcast text extraction
@ocr
Vision Scout
Qwen3-VL-8B frame analysis
@scout
Roster Keeper
Athlete identification & matching
@roster
Clip Director
Clip boundary detection & packaging
@clipdir
Editorial Judge
Clip scoring 0.0–1.0 by editorial value
@judge
Quality Auditor
Quality audit & validation
@auditor
Training Bot
QLoRA training data collection
@trainer
See Full Agent Profiles →

Ready to see AI-powered broadcast analysis in action?

YUHEX is currently invite-only. Request access and we'll review your application within 48 hours. Or explore the documentation to learn how the 11-stage pipeline transforms raw footage into structured intelligence.

Request AccessRead the GuideMeet the Team