№ 02 11 techniques Featured
GAME DEVELOPMENT
Behemoth's game studio: certified slot engines, AI asset pipelines and an LLM generator that turns a brief into a playable game.
01 — Brief
Overview
One studio pipeline covering five shipped game lines — from mathematically certified slot engines to AI-generated prototypes. Go services drive the game logic and fairness auditing, Python pipelines generate art and audio with Gemini and ElevenLabs, and Vue/React frontends render the results through PixiJS, GSAP and Babylon.js on desktop and mobile.
02 — Approach
Challenge & Solution
The Challenge
Game production demands three hard things at once: mathematically verified fairness with a certified payout target, high-frame-rate animation that stays smooth on mobile canvases, and asset/spec production fast enough to prototype new titles in days rather than weeks.
Our Solution
A layered studio pipeline. Clean-architecture Go backends (Fiber + gRPC, GORM, PostgreSQL sessions, Redis state) run the game math with crypto/rand provably-fair RNG validated by Monte-Carlo RTP simulation. Vue 3 + PixiJS 8 and React + Babylon.js frontends handle rendering, with GSAP timelines, Pinia/Zustand state and Howler.js audio tuned for mobile. Python pipelines use Gemini for tiles, glyphs and backgrounds and ElevenLabs for sound, while an LLM spec generator with versioned prompt templates turns natural-language briefs into playable, lifecycle-tracked prototypes.
03 — Capabilities
Key Features
04 — Stack
Technical Architecture
Frontend
- Framework
- Vue 3 / React
- State
- Pinia / Zustand
- Styling
- PixiJS + GSAP canvas, Babylon.js 3D, Tailwind CSS
- Testing
- Vitest
Backend
- Runtime
- Go & Python
- Framework
- Fiber + gRPC / FastAPI
- Database
- PostgreSQL
- Caching
- Redis
Deployment
- Hosting
- Google Cloud Run / Docker Compose
- CI / CD
- GitHub Actions + Make automation
- Monitoring
- Cloud Monitoring
05 — Impact
Results & Impact
Certified 96.92% RTP validated by Monte-Carlo simulation, with provably-fair RNG and a full audit trail
Playable prototypes generated from a natural-language brief via versioned LLM spec pipelines
AI asset pipeline (23 image + 11 sound steps) cuts manual design work across every title
Smooth high-frame-rate PixiJS rendering with animation timing tuned for mobile devices
3D cascade prototype delivered in a short timeframe with accessibility-aware motion handling
06 — Frames