Powered by Amazon Bedrock & Claude AI

The AI Engine
Behind Our Games

Processing 14.9 billion tokens daily across 3 core AI scenarios with 72% cost optimization

Built on Claude Sonnet 4.6 and Haiku 4.5, our platform delivers enterprise-grade AI for coding productivity, multilingual customer service, and intelligent game operations — all running on AWS infrastructure with smart model routing.

7 Bedrock Agents
99.9% uptime SLA
Enterprise security
14.9B
Tokens/Day
AI inference scale
$200K
Monthly MRR
Platform revenue
72%
Cost Saved
vs base pricing
200+
Engineers
Using AI daily
Three Core Scenarios

AI Capability Overview

Our platform serves three distinct AI workloads, each optimized for its unique performance and cost requirements.

AI Coding & Dev Productivity

200+ engineers across Shenzhen · Singapore · Bangalore

Model
Claude Sonnet 4.6 (complex generation) + Haiku 4.5 (fast completion)
Tech Stack
Cocos Creator (TypeScript), Unity (C#), Go, Python
Smart Code Review
PR review time: 4.2h → ≤1.5h (↓64%)
Bug Diagnosis
Localization time: 3.5h → ≤1h (↓71%)
AI Code Generation
IDE completion for VS Code / JetBrains
Auto Test Generation
Unit & integration test scaffolding

Multilingual AI Customer Service & Localization

12 countries · 10+ languages · Real-time conversations

Model
Sonnet (complex dialogue/localization) + Haiku (moderation/QA)
Tech Stack
Real-time chat, content pipeline, moderation system
Intent Recognition
72% (Qwen) → 94%+ (Claude)
Auto Resolution Rate
≥70%, response <3s
Content Moderation
85 items/sec, 94.2% accuracy
Cost Savings
$150K/mo → ≤$60K/mo (↓60%+)

Game Operations AI Agent & Analytics

Intelligent operations across all game titles

Model
Haiku 4.5 (high throughput) + Sonnet (deep analysis)
Tech Stack
Agent framework, analytics pipeline, recommendation engine
Card Balance Analysis
Adjustment cycle: 3 weeks → ≤3 days (↓86%)
Anti-Cheat System
False positive: 5% → 0.5% (↓90%)
Operations Copilot
Analysis: 3 days → 5 min (NL→SQL→Report)
NPC Dynamic Dialogue
3M calls/day, <800ms, 10+ languages
Personalized Push
Open rate: 12% → 18%+
Churn Prediction
Accuracy ≥75%
Benchmark Results

Claude vs Qwen Comparison

April 2026 blind test across 1,200 samples and 450 code tasks. Claude consistently outperforms across all scenarios.

Test Scenario
Claude Sonnet 4.6
Qwen-Max
Gap
Multilingual Intent Recognition (6 langs, 1200 samples)
92.6%
73.7%
+18.9pp
Code Generation Accuracy (450 tasks)
84.8%
61.0%
+23.8pp
Long-Context Bug Diagnosis (20 cases)
85.0%
35.0%
+50.0pp
Content Moderation Throughput (10K items)
85/s, 94.2%
45/s, 78.5%
+89% throughput
Card Balance Reasoning (5 cases)
4.2/5
2.8/5
+1.4 pts
Weighted Overall Score
92.3
56.5
+35.8

Key Takeaway: Claude achieves a weighted overall score of 92.3 vs Qwen's 56.5 — a +63% improvement that directly translates to better player experiences.

Infrastructure

Technical Architecture

A 5-layer architecture designed for high throughput, low latency, and cost efficiency at scale.

Access Layer
CloudFront + ALB (Dual AZ)
Routing Layer
GameAI Hub — Smart routing (95% → Haiku, 5% → Sonnet)
AI Layer
7 Bedrock Agents + Knowledge Bases (RAG) + OpenSearch
Cache Layer
ElastiCache Redis (response cache hit rate 45%+)
Data Layer
Aurora MySQL + S3
Dual AZ Deployment
High availability across availability zones
Smart Routing
95% requests to Haiku for cost efficiency
45% Cache Hit Rate
Redis response caching reduces AI calls
Cost Efficiency

72% Cost Reduction

Smart optimization strategies that reduce our AI infrastructure costs from $723K to $200K per month without sacrificing performance.

Cost Breakdown

Base Price (no optimization)$723K/mo
Prompt Cache (Sonnet 45% hit, Haiku 78% hit)
ElastiCache response cache (45% hit rate)
Smart model routing (95% requests → Haiku)
Batch processing for non-realtime workloads
Actual MRR~$200K/mo
Total Savings: $523K/month (72% reduction)

Optimization Strategies

Intelligent Model Routing

~60% savings

GameAI Hub routes 95% of requests to Haiku 4.5 ($0.25/MTok) and only escalates complex tasks to Sonnet 4.6 ($3/MTok).

Multi-Layer Caching

~25% savings

Prompt caching reduces repeated context costs. Redis response cache eliminates redundant AI calls entirely.

Batch Processing

~15% savings

Non-realtime workloads (analytics, reports, content generation) run in batch mode at 50% discount.

Token Optimization

~12% savings

Structured prompts, response compression, and context windowing minimize token usage per request.

Roadmap

Project Milestones

A phased rollout strategy that minimizes risk while maximizing value delivery at each stage.

Phase 0—
2026/05

AI Readiness Assessment

Phase 1 MVP~$50K
2026/06-07

AI Coding + India Customer Service

Phase 2 Expansion~$150K
2026/07-09

Full-language CS + Moderation + Anti-Cheat

Phase 3 Full Scale~$200K
2026/10-11

All scenarios live + optimization

Build With Our AI Platform

Whether you're looking to integrate AI into your game operations or explore partnership opportunities, let's discuss how our platform can accelerate your goals.

Contact UsSee Our Games