Bianca Starling

AI Discovery System

Hybrid discovery architecture combining semantic retrieval, lexical precision, and conversational orchestration for learning intent resolution.

AIDiscoveryRetrievalMCP

System Overview

This system improves discovery quality for natural-language learning queries by combining retrieval methods and grounding responses in real catalog entities. It matters because keyword-only discovery misses intent, especially for learners who cannot name the exact class or taxonomy term.

Architecture Flow

Learner Intent → Hybrid Retrieval → Grounded Ranking → Conversational Response → Learning Selection Loop

Core Components

  • Intent capture layer: collects and normalizes natural-language discovery prompts.
  • Semantic retrieval layer: matches user intent to catalog meaning, not only literal terms.
  • Lexical retrieval layer: preserves precision for exact terms, titles, and named entities.
  • Orchestration layer: combines signals and returns grounded conversational recommendations.
  • Context protocol layer (MCP): provides structured platform context for safe and relevant AI behavior.

Behavior Loops

  • Learner asks in natural language -> grounded recommendations improve -> class starts increase -> intent signals improve future retrieval.
  • Low-confidence queries trigger fallback behavior -> trust is preserved -> users continue discovery instead of abandoning.

Metrics That Matter

  • Discovery relevance
  • Search to click-through
  • Time to first meaningful result

System note: quality depends on retrieval grounding and explicit fallback behavior, not chatbot UI alone.

Projects

AI Strategy: Semantic Search, Conversational Discovery & MCPSkills Gap AI EngineHeadless CMS & Content Infrastructure

Strategy Notes

AI Discovery Guardrails for Creative Learning

Deep Dives

Query Rewrite Quality Probe

Probe to test whether lightweight query rewriting improves retrieval quality for long, ambiguous learner prompts.