What I’ve built

Eight production AI agents. These aren’t experiments — they run continuously, doing real work for real campaigns. I built them because, as a director, I knew exactly which parts of production were tedious and which parts were creative. I wanted to automate the first.

Story Engine

Multi-agent video storytelling. Seven specialized agents orchestrated through a state machine, with human approval checkpoints between phases. LangGraph, ChromaDB, Claude API, Whisper, Pydantic, Typer, FCPXML/EDL.

• 16,000+ soundbites in the RAG database
• 95% quote accuracy rate
• 7 agents orchestrated
• 3 export formats supported

Media Toolkit Pipeline

Automated organization and AI-powered analysis of large video and photo libraries. Scene analysis, contact sheets, searchable HTML catalogs. Claude Vision, YOLO, FFmpeg, SQLite, Pydantic.

• 86,342 files scanned
• 40,256 files organized
• 2,242 contact sheets generated
• 34 HTML catalogs created

Morning Briefing Agent

A 6:30 AM daily briefing on Slack — emails, calendar events, and news, prioritized. GPT-4o, Gmail API, RSS, Slack. 20 RSS feeds monitored, 20 VIP contacts tracked.

Calendar Assistant

Voice-first scheduling. Whisper transcription + GPT-4o + Google Calendar. Speak a request, the agent puts it on the calendar.

Task Planner

ClickUp + Claude. Weekly planning automation with natural language queries and ICS calendar exports. ~40 hours per week routed.

Pics.io Router

Polls the asset inbox every five minutes, routes uploads to the right collection based on metadata. 28 collections mapped. Docker, REST, Pics.io API.

Notes Capture Agent

Voice-to-Obsidian. Speak a thought; the agent transcribes it, tags it, files it in the right folder. Gemini, Slack, Obsidian. 7 folders routed.

Finance Analysis

Transaction analysis and spending pattern visualization. Python, Pandas. In development.

Tech I work with

LangGraph, Claude (Anthropic), GPT-4o, Gemini, ChromaDB, Whisper, YOLO, FFmpeg, Pics.io, Docker, Python, Pydantic, Typer, MCP servers.