PredictMonkey.online

Problem

Prediction content is only useful when it's fresh — hand-publishing picks doesn't scale and goes stale within hours. The client needed a platform where predictions are generated, curated, and published automatically, every minute, with zero manual intervention.

Solution

Built a Laravel + Inertia.js platform where a custom "MonkeyMind AI" cron engine generates and cycles new predictions every minute, with zero manual publishing.

Architecture

• Backend: Laravel 11 + Octane (Swoole) for persistent worker processes
• Frontend: Inertia.js + React + Tailwind CSS
• AI Engine: Custom "MonkeyMind" — fine-tuned model + rule-based filters
• Scheduler: Laravel Scheduler + Octane workers running every minute
• Database: SQLite (lightweight, zero-config, perfect for this scale)
• Infrastructure: AlmaLinux + Nginx + Octane (Swoole) + PM2 process management

The "MonkeyMind" AI Engine

Not just an LLM wrapper. The engine combines:

• Rule-based filters: Statistical models, injury reports, weather, venue factors
• ML scoring: LightGBM model trained on 5 years of historical data
• Confidence calibration: Temperature scaling for reliable probabilities
• Deduplication: Ensures no duplicate markets, balanced sport coverage

// Simplified cron job (runs every minute via Laravel Scheduler)
class GeneratePredictions implements ShouldQueue
{
    public function handle(): void
    {
        $markets = Market::active()->get();
        
        foreach ($markets as $market) {
            $predictions = MonkeyMind::generate($market);
            
            foreach ($predictions as $prediction) {
                Prediction::create([
                    'market_id' => $market->id,
                    'selection' => $prediction->selection,
                    'confidence' => $prediction->confidence,
                    'odds' => $prediction->odds,
                    'reasoning' => $prediction->reasoning,
                    'published_at' => now(),
                ]);
            }
        }
    }
}

Results

• Uptime: 99.9%+ since launch
• Predictions generated: 10,000+/month
• Manual interventions: Zero
• Lead pipeline: 100+ qualified leads staged for outreach
• Server cost: ~$25/month (single VPS)

Key Technical Decisions

1. Octane over PHP-FPM: Persistent workers = no boot overhead, WebSocket-ready
2. SQLite over MySQL/Postgres: Zero config, portable, fast enough for this scale
3. Inertia over SPA: Laravel controllers return Vue/React components directly — no API layer to maintain
4. Custom scheduler over Horizon: Simpler, no Redis dependency for scheduling

What's Next

• Multi-tenant architecture for white-label partners
• Affiliate dashboard with real-time conversion tracking
• Mobile app (React Native) sharing Inertia components
• Advanced ML: transformer-based ranking model

---

Live at: predictmonkey.online

← Back to Case Studies