Ingest Factory
Acquire, stage and organise incoming match footage for the processing pipeline.
RISE Broadcast Platform
From Detection to Intelligent Highlight Creation
Raw footage to searchable, reviewable, exportable highlights — one connected workflow, operator-controlled at every step.
0.79
Macro F1
6,943
Labelled clips
38
Event classes
Built from live production pressure
A clean engine, built on operational reality.
RISE connects ingest, detection, tracking, event recognition, operator review, archive intelligence and highlight creation inside one reviewable workflow — keeping the creative decisions where they belong.
One ingest, every output.
Detection
Scene understanding identifies players, objects and spatial relationships across the match.
Tracking
Player and object continuity maintained across the full match timeline.
Event Recognition
38 event classes mapped for searchable, filterable football intelligence.
Operator Review
Every prediction carries confidence and attribution — visible before anything reaches production.
AutoLabel
Reviewed footage feeds assisted labelling and accelerated training cycles.
Highlight Creation
Approved intelligence composed into exportable, structured highlight output.
Platform in action
Live product surfaces, not placeholder mockups.
Predictions archive, operator review, detection and catalogue views — one connected operating system.
Predictions Archive
Every clip. Every confidence score. Every decision.
The archive surfaces every detected event with status, confidence and attribution — making the full prediction history reviewable, filterable and auditable before anything reaches production.
92%
Archive done
5,389
Clips processed
Operator Review
Human control at every stage.
Every automated output passes through a review layer where operators can accept, reject or adjust predictions with full timeline context. No clip reaches production without explicit operator sign-off.
1,376
Review actions
38
Event classes
AutoLabel
Reviewed footage feeds the next training cycle.
Once an operator approves detections, AutoLabel packages them for batch training — closing the loop between review and model improvement automatically. The archive compounds into a permanent operational advantage.
Catalogue Review
The full archive, filterable and searchable.
Every processed match, every event class, every confidence band — surfaced in a structured catalogue that makes re-use, audit and highlight extraction operational rather than manual.
ML progress
Model quality climbs with every review cycle.
0.79
Macro F1 — May 2026
Recognition baseline sharpens as reviewed footage compounds into stronger labels and faster training cycles.
Macro F1 trajectory
Review-led model gain
The data flywheel
Every operator correction feeds training.
More reviewed footage means stronger labels, sharper models and faster future review. The archive compounds into a permanent operational advantage.
01
Review
Operator approves detections
02
Label
Clips packaged for training
03
Train
Models improve automatically
End-to-end workflow
Nine stages. One connected pipeline.
01
Acquire
02
Ingest
03
Detect
04
Track
05
Label
06
Train
07
AutoLabel
08
Highlight
09
Publish
Investor materials
Review the technical proof, execution roadmap and pilot plan.