MUDRAIS: Semantic Matchmaking Engine

MUDRAIS (Multi-User Dynamic Roleplay AI System) solves a fundamental problem in online communities: finding truly compatible collaborators. Traditional bots rely on keyword filters that miss nuance — narrative tone, writing style, thematic compatibility. MUDRAIS replaces that with a semantic matchmaking engine. When a user registers, their free-text input enters a multi-agent AI pipeline running on AMD MI300X via AMD Developer Cloud. A Gatekeeper agent (OSS 20B) extracts structured fields from natural language. A Context Optimizer agent (OSS 120B) generates a semantically dense representation tuned for embedding. The nvidia/llama-nemotron-embed-vl-1b-v2 model then converts it into a 2048-dimensional float vector stored in Qdrant. The matchmaking engine runs cosine similarity search inside Qdrant, applying hard payload filters for "red lines" (absolute blockers) and strict guild-level tenant isolation — users from Server A mathematically cannot appear in Server B results. A secondary scoring layer applies soft penalties (yellow lines, timezone mismatch) on top of raw cosine distance to produce a final compatibility score. The system is Discord-agnostic by design. Discord is just the transport layer (slash commands, webhooks). The entire semantic pipeline — profile ingestion, multi-agent optimization, vector indexing, and partner search — lives in a clean Domain-Driven Design architecture on Laravel. The same engine can power Slack bots, web forms, or REST APIs without touching domain code. Each community is an independent tenant with isolated Qdrant payloads, its own archetype configuration, and its own LLM prompts — true B2B multi-tenancy at scale.