Drop Photos or MCP Prompt
User drops photos into the web portal or instructs Claude via natural language MCP tools.
Automated AI ingestion pipeline and Model Context Protocol (MCP) server for CUHK Canto Digital Asset Management, featuring 5-skill agent, spatial college derivation, and binary IPTC exiftool embedding

5-skill autonomous agent deriving spatial college taxonomy with binary IPTC injection and remote MCP tools.
Modular 5-skill agent on Vertex AI with spatial college derivation, exiftool IPTC embedding, and remote MCP tools.
Automated landmark identification across 170+ buildings and 40+ landmarks with spatial college derivation.
Manually cross-referencing 170+ buildings, 40+ landmarks, and 10 album categories made campus photo ingestion error-prone.
Modular 5-skill AI agent on Vertex AI with spatial majority-vote college derivation and remote MCP tools for AI clients.
CASE STUDY ROLE: LEAD INTEGRATION ENGINEER & AI AGENT ARCHITECT
Xavier designed the modular 5-skill AI agent where each tagging instruction (albums, buildings, landmarks, descriptions, tags) is stored in a clean markdown file.
He developed a spatial majority-vote algorithm that deduces which of the 9 colleges a photo belongs to, used exiftool to stamp IPTC metadata into the image binaries, and built an MCP server so AI assistants like Claude can search and upload photos using plain English.
User drops photos into the web portal or instructs Claude via natural language MCP tools.
Gemini identifies 170+ buildings, 40+ landmarks, 10 albums, and writes 20 search tags.
Deduces college affiliation via majority vote and embeds permanent IPTC metadata via exiftool.
Uploads binary to Canto's S3 CDN, polls for asset creation, and fills all metadata fields.