poqpoq World is an AI-first metaverse where companions reason about relationships, remember conversations, and perceive the 3D world around them. Every NPC maintains its own memory, personality, and awareness of the environment it inhabits.
This documentation explains the engineering behind real-time, personalized AI at scale — from semantic memory retrieval to adaptive perception systems that run within the tight latency budgets of a live virtual world.
Player Guide
AI Architecture
Chapter 1
AI Foundations
How companions think: tiered intelligence, memory architecture,
and the five core principles that shape every design decision.
Chapter 2
Memory & Embeddings
Semantic memory, significance scoring, hybrid search,
and the three layers of decay that keep recall human-like.
Chapter 3
Attention & Perception
How companions sense the world: adaptive sampling,
deduplication, and spatial awareness in real time.
Chapter 4
The Model Philosophy
Why curated context with a capable model outperforms
massive context with a smaller one.
Chapter 5
Vector Search & Optimization
FAISS, approximate nearest neighbors, and the complete
optimization pipeline from embedding to retrieval.
Chapter 6
Case Study: The Identity Crisis
How hardcoded identity in AI prompts caused a silent failure
across every companion — and the dynamic identity fix.
Chapter 7
Creative Agency
When AI companions start creating on their own: emergent behavior,
pattern recognition, and what it means for the future of AI.
Chapter 8
RAG Architecture
Retrieval-Augmented Generation: why RAG beats fine-tuning,
hybrid search design, and knowledge seeding at scale.
Chapter 9
From Knowledge to Agency
The leap from retrieval to action: how companions learn to
manipulate the world through natural language commands.
Chapter 10
Dungeon Mouths & The Infinite Quest Loop
Object donation, sacrifice mechanics, seed-based reality,
and the economic flywheel that turns building into questing.
Chapter 11
Dungeon Instances
Private party adventures: instanced PvE zones with NPC spawning,
checkpoints, loot distribution, and difficulty scaling.
Creator Guide
Community
Who is this for?
Developers
Building AI-powered virtual worlds and need practical patterns
for memory, perception, and real-time reasoning.
Researchers
Interested in applied AI architecture — how embedding search,
tiered models, and adaptive systems work under production constraints.
Educators
Teaching applied AI and ML systems with a concrete,
end-to-end case study from a shipping product.