Let the Grappler Eyes
analyse your video.

RollIQ is a computer-vision system built to extract everything that matters from your grappling sequences. This page shows the pipeline most recent, unedited output.

See the dashboard mockup →
What we're trying to solve

Feedback is the most powerful improvement factor

The problem

Your rolls go unreviewed

A master coaching in front of hundreds of practitioners on the mats

Unfortunately, in martial arts your coach or master can't see all your mistakes, because there are dozens of other practitioners. On top of that, if you don't explicitly ask for advice, your technique can plateau for several months.

The best coaches can only watch a few rolls per member. The sequences that work well for you and the positions that you should work on go unreviewed. However, just like in any other sport, the best athletes are the ones who can learn from EVERY single training session.

What the Grappler Eyes can do

Automated analysis of your grappling sessions

The solution here is not to replace your master. The solution is to bring him the right material, so he can finally offer you the best guidance.

  • Record as many rolls as you want for a precise understanding of your game.
  • Upload them — AI makes the highlights and gives you the full breakdown.
  • Get reminded of the techniques you should work on, with top-quality tutorials recommended directly through the app.
  • Identify the positions you statistically often get stuck in, important techniques missing in your game, and compare your game style with your gym buddies.
  • Analyse the game of your next opponent.
Grappler Eyes — live output

Current Analysis Capabilities

I've been building RollIQ for 5 months now. I'm both the engineer and the guy who trains. I'm building this app, and using it myself to prepare for my first competition because I noticed that whenever i'd record my rolls and review them, I'd progress so much faster. I'm convinced that every athlete willing to prepare his gameplan deserves a proper breakdown of every roll.

Everything below is real output. It's running on an unedited IBJJF Worlds match. Some of it already works well, some definitely not yet, but more importantly, I built it in a way such that it acknowledges uncertainty rather than guessing for the best. Because a system that fakes confidence is worse than useless for something you'd actually use to prepare a gameplan.

— built by a grappler for grapplers
🎥
Personal video coming soon
Drop a file at media/personal_intro.mp4 and this placeholder is replaced automatically.
Athlete 1 (confident ID) Athlete 2 (confident ID) ID lost (honest uncertainty) Referee (flagged, never confused for a competitor)
Textbook open guard, both athletes tracked Open guard position continues
✓ Working now

Textbook open guard

Athlete 1 plays open guard against a standing Athlete 2. Both stay clearly identified the whole time.

Sustained tracking through a leg entanglement Submission finish, both athletes confidently tracked
✓ Working now

Back mount → triangle → submission

Athlete 1 passes to back mount, locks a triangle, and finishes the match. Both athletes stay identified all the way through the tap.

Merged detection during entanglement Overlapping candidate boxes during entanglement
→ Getting there

Bodies merge into one detection

The two athletes compress into one detection box. The system shows this honestly instead of guessing who's who.

Referee stepping in front of athletes Referee overlapping athlete during check
→ Getting there

The referee walks directly into frame

The referee crosses in front of both athletes. He stays flagged separately and is never confused for a competitor.

Under the hood

How Grappler Eyes actually works

Five stages, each one a real, separate piece of the pipeline.

01

Video in

Full, unedited match footage

02

Detection + Pose

Every athlete, every frame

03

Tracking

Consistent identity, honestly reported

04

Position & technique

Classified and timestamped

05

LLM narration

Plain-language breakdown

Where this goes with more data

What improves from here is mostly about collecting more data. With enough annotated data, we'll be able to reconstruct positions in 3D and offer detailed technique analysis — going from today's ~15 broad position classes to recognizing basically any variation of any position, and eventually giving feedback on the techniques themselves.

  • A full position timeline for the whole match — every second accounted for.
  • Automatic scoring: takedowns, guard passes, mount, back control, and sweeps tallied as they happen.
  • A plain-language report per roll — what worked, what didn't, and what to drill next.
  • Patterns across many rolls over time — recurring weak positions, techniques you keep running into.

If I can show enough people are interested, I can get the funding to collect that data. Check below.

This is a mockup of the finished RollIQ product — not live output. See real pipeline output →

Upload & Queue

Drop in every roll you film. Grappler Eyes processes them in the background and finds patterns across your whole training history.

Drop a video here, or browse files
MP4 · up to 2 hours · processed in the background

Processing queue

gym_roll_aug02.mp4
Processing · 62% · ~1m40s left
comp_finals_jul28.mp4
Done · report ready →
comp_prep_session.mov
Done · report ready →
open_mat_jul21.mp4
Queued

Insights from your last few sessions

6 successful double-legs recorded last week
Your takedown entries are trending up across the last 3 sessions.
3 losses by kimura recorded
This escape might help — Kimura Escape from North-South →
Best of your last competition
Auto-cut from your 3 matches — the highlight reel below.

Auto-generated highlight reel

Concatenated from your last 3 uploaded sessions, ranked by scoring events and clean submissions.

Help build this

I'm building RollIQ alone — no team, no funding, just several months of nights and weekends so far, training and coding on the same days. If you train — any discipline, any level — and want to help shape what this becomes, I'd love to hear from you. I'll reach out about specific footage, feature ideas, or feedback on what actually matters to a coach or competitor.

Buy me a coffee →
🤝

Help make RollIQ real

I'm building this alone, nights and weekends. The more people signed up, the easier it is to make the case for funding the data collection this needs. Leave your email and I'll keep you posted as it grows.

No spam — just progress updates and early access.
Prefer the full form? →