Problem
Judging, broadcast stats, scouting, and fight review still rely on slow manual workflows. The result is limited data, weak auditability, and missed storytelling value for regional promotions.
Reviewer technical brief
AI command layer for live combat sports.
Fight Judge AI converts fight footage into real-time scoring signals, timestamped audit trails, broadcast overlays, APIs, and FightMentor agentic coaching workflows. The platform is designed around GPU-accelerated video AI workloads, not a generic SaaS wrapper.
Core thesis
Combat sports need live, explainable fight intelligence.
Platform
Fight Judge AI ingests footage, detects combat-specific events, builds a fight intelligence graph, and ships scoring evidence to operators, broadcasts, APIs, and FightMentor agents.
Expansion
ScoreOps creates the event-data wedge. FightMentor expands that graph into coaching, scouting, and gym workflows with cited clips and training recommendations.
System architecture
From camera feed to audit trail.
01 / Capture
Ingest live camera feeds or uploaded fight footage from event operators.
02 / CV inference
Detect pose, contact, strike events, cage position, motion, pressure, and control time.
03 / Event graph
Normalize frame-level signals into fighter, exchange, round, and scoring objects.
04 / Scoring model
Map damage, dominance, duration, and control into explainable round recommendations.
05 / Audit output
Attach score decisions to timestamps, clips, confidence, and human-in-loop review.
06 / Product surfaces
Power broadcast overlays, score reports, APIs, and FightMentor agentic analysis.
NVIDIA acceleration fit
Workloads mapped to accelerated computing.
Low-latency frame analysis for strike, pose, motion, and control-time signals.
Combat-specific recognition under occlusion, rapid exchanges, cage contact, and camera motion.
FightMentor retrieves timestamped clips, fighter profiles, metrics, and coach notes with citations.
Model calibration, replay analysis, synthetic review, and continuous scoring improvement.
Venue-side inference for live events where latency and connectivity can block cloud-only systems.
Structured data feeds for overlays, reports, scorecards, and partner integrations.
Moat
Defensibility comes from combat-specific data and review loops.
Data graph
Fight records, fighter profiles, fight videos, event metadata, scoring labels, and timestamped CV evidence form a domain-specific corpus.
Human feedback
Judges, coaches, commissions, and operators create review data that improves model confidence and score explanations.
Operational wedge
Regional events provide the live workflow where FightJudge can create data, prove value, and expand into FightMentor intelligence.
Reviewer demo path
Recommended 15-minute walkthrough.
Show ScoreOps live timeline, Audit Lens evidence behind a 10-9 round, FightMentor Agent Lab with cited clips, and the GPU roadmap for edge/cloud inference and training.