Specialization · Esports competitive intelligence

The gap isn't talent.
It's information.

Top Chinese, Korean, and SEA esports teams have one thing emerging rosters don't: analysts, databases, and scouting reports. They know your tendencies before they drop. You know theirs from gut feel and unrewatched VODs.

Datavyas extracts the same competitive intelligence — automatically, from the tournament broadcast itself. The capability international teams have spent a decade building. Delivered as a service.

18
matches
PMGC 2025 Gauntlet — processed automatically
1,500+
events
Kills and knocks extracted with attribution
~95%
accuracy
On clean broadcast feed. 100% on local feeds.
150h → 0
manual work
What used to take an analyst a month
The problem

Every match generates thousands of data points. No team captures them.

A single PMGC match: 64 players, 16 teams, hundreds of engagements. Manual extraction takes 3–4 hours per match. Studying 50 matches to find real patterns: 150+ hours of analyst time. So almost nobody does it — and the teams who do keep their work private.

Top international teams
  • Dedicated analyst staff studying every VOD
  • Multi-tournament databases on every opponent
  • Pre-tournament scouting reports per matchup
  • Quantified playstyles, timings, target priorities
Most emerging-region rosters
  • Watch VODs occasionally, rewatch rarely
  • Compete on instinct against teams with files on them
  • Discover an opponent's tendencies mid-tournament
  • Lose to preparation, not mechanics

The gap is preparation. The lever is automation. That's what we built.

What we extract

VOD in. Structured intelligence out.

Drop a tournament broadcast into the pipeline. Get back a clean, queryable database of every event, every player, every match — ready for analysis.

Kill feed

Every engagement

Killer, victim, action type (knock vs kill), and timestamp. Weapon probability where detectable.

~95% on broadcast
Standings

Match rankings

Final ranks, placement points, elimination counts, total points — captured across the full match.

~95% automatic
Survival

Elimination order

Players remaining over time. Teams remaining over time. The exact frame each team was wiped.

~95% automatic
Engagements

Who fights whom

Derived from kill-feed flow: matchup frequency, rivalries, third-party opportunities.

Derived
What the data tells you

Six questions your IGL has been guessing on.

Each one answerable from extracted data alone. Each one a decision you make every match.

01Question

Should we take this fight?

Head-to-head intelligence

You vs Team X across 30 matches: you killed them 18 times, they killed you 27. 40% win rate. Avoid. Third-party or disengage.

02Question

Who do we need to watch?

Danger player identification

One player accounts for 45% of Team X's total kills. Prioritize him in any engagement. The other three are secondary.

03Question

When are they dangerous?

Timing patterns

Team SOUL: 18% of kills early game, 52% late game. They're a late-game team. Fight them in Phase 1–2 or not at all.

04Question

How do they actually play?

Playstyle classification

Aggressive early, passive late, frag hunters, placement players, balanced — every team scored on the matrix from real data.

05Question

Who do they hunt?

Targeting patterns

Team DRX takes 23% of their kills from SOUL. That's a rivalry, not a coincidence. Position to third-party.

06Question

What does the full opponent look like?

Scouting reports

A one-page profile per opponent: playstyle, key players, head-to-head, when to engage, when to avoid. Generated, not written.

Proof — already built

Tournament video in. Match database out.

The pipeline runs end-to-end today. It has processed real tournament footage from PMGC 2025 Gauntlet — 18 matches, 1,500+ kill events, full standings, full survival data — fully automated.

Pipeline · Broadcast → database
  1. 01
    Frame extraction
    ffmpeg sampling at 3 fps. Live and replay modes.
  2. 02
    Region detection
    Kill-feed slots, team HUD, points table — cropped per frame.
  3. 03
    Event classification
    Custom ResNet-18 distinguishing Kill, Knock, and noise frames.
  4. 04
    Player resolution
    PaddleOCR + fuzzy roster matching defeats OCR noise on stylized player names.
  5. 05
    Attribution engine
    Game-rule encoded: kills credited to the knocker when a teammate finishes the target. Team wipes detected. Ranks derived from wipe ordering.
  6. 06
    Scouting database
    Clean structured output, queryable for head-to-head, danger players, timing, playstyle, targeting, full opponent reports.
Tournament
PMGC 2025

Gauntlet stage · all 3 days · 18 matches

Events extracted
1,500+

Kills, knocks, and standings — fully attributed

Manual baseline
150+ hours

What this replaces. Per 50 matches. Per team.

The architecture is broadcast-agnostic. PUBG Mobile is proven. The same pipeline generalizes to other kill-feed and HUD-driven games.

The edge

Why this works where nothing else does.

Broadcast is the source

No publisher API. No cooperation required. If the tournament has a VOD, we have the data. That's the entire competitive moat for esports analytics outside the top-five publishers.

Rules-aware analytics

Generic counters miss what matters. We encode game logic: knocker-credit on finishes, team wipes from per-player state, placement vs frag points. The numbers reflect how the game actually counts.

Honest about scope

WHO killed WHOM, and WHEN — today, with high accuracy. WHERE on the map — Phase 2. We tell you what we can answer and what we can't. No vapor.

Roadmap

Today we answer WHO + WHEN. Next we add WHERE.

Now
Shipped

Combat intelligence

  • Kill-feed extraction
  • Match standings & survival
  • Head-to-head & playstyle
  • Danger players & timing windows
  • Generated scouting reports
Next
In development

Location intelligence

  • Minimap parsing
  • Rotation tracking
  • Zone-circle detection
  • Map-specific patterns
  • Position-relative analysis
Future
Planned

Real-time & predictive

  • Live integration during scrims
  • POV-level breakdowns
  • Predictive rotation modeling
  • In-tournament intelligence feed
Who it's for

Built for the side that doesn't already have analysts.

The teams who need this most are the ones who can't afford what the top-five rosters spend on it. We close that gap.

Primary

Esports teams

Rosters competing in regional and international tournaments. Pre-tournament scouting, post-match review, opponent files.

Org

Esports organizations

Multi-roster orgs needing shared intelligence infrastructure across titles and squads.

Broadcast

Tournament organizers

Live structured stats for broadcast augmentation, narrative graphics, and post-event reports.

Markets

Odds & fantasy

Real-time event streams for live betting markets and fantasy pricing — derived from public feeds.

Pilot program

We process your next tournament.
You bring the talent.

A bounded pilot for teams ready to compete on preparation, not just mechanics. We process the relevant tournament VODs, build the opponent database, and deliver scouting reports for your next bracket.

Step 1
Tournaments processed

3 months of relevant VODs ingested end-to-end

Step 2
Scouting reports

Per-opponent profiles for your next major tournament

Step 3
Dashboard access

Live access to head-to-head, players, timing, playstyle

Step 4
Iteration loop

You name the insights that matter; we prioritize extraction

“In any competition, the team with better information makes better decisions. Better decisions compound into better results. This isn't theory — it's math.”