AAII — Phase 2 - MCP Server, Jupyter Notebooks

Portfolio Snapshot

Phase 2 extends the AAII platform with an AI tooling layer, exposing the deployed inference and semantic search capabilities through MCP (Model Context Protocol), notebooks, and a comparative UI. This phase focuses on LLM-facing interfaces, evaluation workflows, and AI system ergonomics, rather than infrastructure.

What it proves

• Ability to expose AI capabilities as standardized tools for LLM clients (MCP), not just HTTP APIs.
• Practical understanding of when to use direct APIs vs tool-mediated calls.
• Experience designing AI systems that are inspectable, testable, and evaluation-friendly.
• Comfort working across API, tool protocol, UI, and notebook-based analysis layers.

Key components

• Minimal MCP server delegating to live AAII API endpoints (/model, /embed, /search).
• MCP tool surface for semantic search and embeddings over a real JSON corpus.
• Blazor UI demonstrating side-by-side:
  • direct API calls
  • MCP tool-layer calls
• Jupyter notebooks for:
  • embedding inspection and similarity analysis
  • semantic search validation and ranking behavior
• CPU-friendly transformer embeddings with deterministic outputs and cached document vectors.

Operational notes

• MCP layer is intentionally thin (no duplicated business logic).
• All tooling runs against the same deployed API used in Phase 1.
• Notebooks and MCP tools are designed for learning, debugging, and evaluation—not production automation.

Verification

Visual evidence of project