Eagle Esports

01 Project Overview

One of the key problems addressed by the platform is the manual processing of match-result screenshots. Tournament staff traditionally need to read screenshots, identify players and teams, enter kills and rankings, calculate points, and update standings manually.

The Solution The AI Magic Point Scanner automates this result-processing workflow by extracting match information from screenshots, matching the extracted players with registered teams, and generating editable result entries for staff verification before the data is saved.

02 My Role

My responsibilities included end-to-end implementation of the platform and the AI workflow:

Designing and developing the application workflows for players, staff, administrators, payment teams, and public match-room users.
Developing PHP/MySQL APIs for authentication, team management, registrations, match points, proofs, notifications, payments, and roster management.
Building the AI Magic Point Scanner workflow, including request handling, validation, structured-output parsing, caching, player/team matching, and manual review.
Connecting screenshot processing with the existing tournament points and results system.
Implementing controls for multiple screenshots, uncertain names, temporary rosters, manual corrections, and review states.
Developing responsive interfaces using HTML, CSS, Tailwind-generated styles, and JavaScript.
Implementing security and operational controls including role-based authorization, upload validation, rate limiting, transactions, cache locking, and audit logging.

03 System Architecture

The application uses a web frontend backed by a centralized PHP API and MySQL database. Browser requests communicate with named API actions, while the AI scanner processes screenshots server-side through a multimodal AI endpoint.

Player / Staff / Admin
       │
       ▼
HTML + CSS + JavaScript UI
       │
       │ JSON / Multipart Requests
       ▼
PHP API (api/index.php)
       │
       ├─────────────┼─────────────┐
       │             │             │
       ▼             ▼             ▼
   MySQL       Multimodal AI    Firebase
  Database         API            FCM
       │             │             │
       │             ▼             │
       │       Image Processing    │
       │       & Data Extraction    │
       │             │             │
       └─────────────┴─────────────┘
                     │
                     ▼
              Match Results
                     │
                     ▼
              Leaderboards

04 Technology Stack

A pragmatic, production-oriented stack combining classic web engineering with modern multimodal AI capabilities.

Frontend

HTMLCSSJavaScriptTailwind-generated CSSRemix Icon

Backend

PHPPHP mysqliPHP cURLPHP GD

Database

MySQLSQL

AI & Vision

Multimodal AI APIQwen vision model (OpenAI-compatible)Structured JSON extractionImage-based text & data recognition

Other Technologies

Firebase Cloud MessagingJWT authenticationBcrypt password hashingNode.js toolingJS/CSS minification tools

05 AI Magic Point Scanner

The scanner supports multiple screenshots for a single scan. The backend validates image content and size, generates a scoped cache key, sends the multimodal request, validates the returned JSON, and presents the results as editable entries for staff review.

Match Screenshot
       ↓
Image Validation
       ↓
Multimodal AI Processing
       ↓
Structured JSON Extraction
       ↓
Response Validation
       ↓
Player / Team Matching
       ↓
Staff Review & Correction
       ↓
Point Calculation
       ↓
Database Persistence
       ↓
Leaderboard Update

AI Processing Pipeline

The scanner uses a server-side multimodal AI request rather than a separate standalone OCR SDK. The current implementation uses the configured Qwen vision model qwen/qwen3.7-flash through an OpenAI-compatible API with non-streaming output and deterministic temperature settings.

Screenshot upload and validationSecure handling of image content and size checks before processing.

Multimodal image processingServer-side request to the configured vision model.

Extraction of player names, ranks, and killsStructured fields parsed from screenshot content.

JSON-only structured outputForced deterministic, non-streaming response format.

Response parsing and validationMalformed or incomplete results are detected and flagged.

Manual review before saving resultsNo AI result is persisted without staff approval.

06 Player and Team Matching

AI-generated player names may contain spelling differences or formatting variations, so the backend performs additional matching against registered players using multiple strategies:

Case-insensitive exact matching.
Case-insensitive containment matching.
Similarity matching using PHP similar_text().
Normalized score calculation.
Duplicate-team prevention.

A minimum normalized score is required before a team is accepted. Unmatched or uncertain results remain available for manual correction rather than being silently accepted.

07 Human Review and Validation

The scanner does not directly publish AI-generated results. Every extraction passes through a verification, review, and correction stage before being saved to the database.

AI Result → Validation → Review → Correction if required → Save

Staff can perform the following actions during review:

Correct player names.
Remove incorrect entries.
Add missing entries.
Review uncertain matches.
Confirm temporary rosters.
Approve results before persistence.

Roster entries below the configured confidence threshold are explicitly marked for review and cannot be saved until corrected.

08 Automated Point Calculation

The scanner extracts kills and placement information rather than directly deciding the final tournament score. The application then calculates points using the configured kill-point value and placement-point template. The final standings are calculated using the stored match results and configured scoring rules.

09 Admin Dashboard

The administration system provides centralized management for tournament operations. Implemented capabilities include:

Tournament and lobby configuration
Match creation and management
Team and registration management
Slot assignment
Room ID and password management
Match result management
AI Magic Point Scanner
Temporary match rosters

Result and qualification management
Winner announcements
Payment administration
Proof management
Notification management
Staff/member role management
Audit logging
Public match-room configuration

10 Database Design

The platform uses MySQL for persistent application data. The primary saved match-result flow connects registrations with map-level point records, while temporary rosters provide a match-specific player snapshot without modifying the permanent team roster.

esports_teams esports_players registrations match_points match_proofs payment_orders session_tokens fcm_tokens scheduled_notifications staff_activity_logs match_temporary_roster

11 Security & Reliability

Bcrypt password hashing
JWT-based player authentication
HttpOnly authentication cookies
Server-side staff sessions
Role-based authorization
Lobby-scoped access control
Login throttling
OTP verification

Prepared SQL statements
Secure image-upload validation
AI request validation
Database transactions
Audit logging
Notification controls
Cache locking

12 Performance & Concurrency

The implementation includes several safeguards for concurrent activity. The system is implemented as a centralized PHP API rather than a distributed worker architecture, so AI processing remains synchronous at request time. Caching and locking are used to reduce duplicate work and unnecessary provider requests.

Database transactions for multi-step operations.
Scoped cache keys based on task, match, images, model, and context.
File locks to prevent duplicate AI requests for the same scan.
Notification batching and scheduling.
Scoped result queries.
Image-size limits and secure image processing.

13 Key Engineering Challenges

Converting complex game screenshots into structured data
Handling multiple screenshots with overlapping information
Matching extracted player names to registered teams
Handling spelling and formatting variations
Preventing duplicate team assignments

Keeping AI results editable and reviewable
Maintaining separate temporary rosters for match-specific situations
Persisting results across multiple maps
Avoiding duplicate AI requests through caching and locking
Maintaining responsive staff workflows for large result sets

14 Project Highlights

The implementation currently supports:

Up to three screenshots per AI scan.
Image validation with per-image and combined request-size limits.
Multiple supported Free Fire maps.
Temporary roster management for match-specific player assignments.
Automatic kill and placement extraction.
Player/team matching.
Manual correction workflows.
Cached AI responses.
Transactional multi-map result persistence.
Real-time notification capabilities.
Secure file-processing workflows.

15 Final Overview

Eagle Esports combines full-stack web engineering with practical AI integration to solve a real operational problem in tournament management. The AI Magic Point Scanner transforms match screenshots into structured player, kill, and placement data; validates the results; matches players to registered teams; allows staff review and correction; calculates tournament points; and persists the final results into the tournament system.

The project demonstrates experience across:

Full-Stack Development AI Integration Computer Vision API Development Database Engineering Authentication Automation Security Production Deployment