George.dev
Back to Projects
Mozisha
AI/ML

Mozisha

Full-Stack & AI Engineer

Next.jsNestJSPostgreSQLpgvectorMCP SDK
Live Site

About the Project

Mozisha (Taste Engine) is an infrastructure platform for human-led AI orchestration. Users capture AI output rejections as reusable 'taste bar' constraints and audit their judgment through a 6-step Judgment Loop. The platform compresses years of experience into verified expertise, generates portfolios from validated work, and enables sharing constraints through a Pan-African Constraint Commons.

Key Highlights

  • Built the Judgment Loop — a 6-step workflow for capturing and encoding human quality standards
  • Implemented semantic search with pgvector for finding similar constraints and patterns
  • Integrated MCP SDK for Claude IDE integration, enabling taste-aware AI assistance
  • Built portfolio generation system from validated judgment data using PDFKit
  • Designed constraint sharing system (Pan-African Constraint Commons) for community knowledge

Technical Challenges

Encoding subjective quality judgment ('taste') into structured, reusable constraints required a novel data model. The semantic search system needed to find constraints that were conceptually similar — not just keyword matches — which led to implementing OpenAI embeddings with pgvector for vector similarity search.