Work that runs

AI systems built to run inside real businesses: knowledge engines, automations, and agents. Each one ends in an outcome, not a demo.

All projects

knowledge · Active

AI is only as useful as what you can put in front of it.

Second Brain / Open Brain

A working personal knowledge OS: an MCP server with semantic search over pgvector, daily capture, processing, and digest pipelines, holding 11,000+ entries across 59 knowledge domains. It runs my daily practice.

11,000+ entries across 59 domains

Knowledge corpus

PythonMCPPostgreSQLpgvector
content · Active

A beautiful ad that does not land its message is a beautiful failure.

Obscura

An automated ad pipeline: a product blurb becomes a multi-scene storyboard, drift-free character renders, and a scored MP4, gated by an audit that measures whether a cold viewer gets the message.

Blurb to finished ad video, end to end

Pipeline

PythonSeedancefal.aiFFmpeg
consulting · Active

The numbers rebuilt, not accepted.

Bid & Investment Screening for an Infrastructure Group

A full priced tender package produced in 3 days for a six-country infrastructure group, plus an investment screen that caught a falsified growth claim before money moved.

Full priced bid in 3 days, against a 4 to 6 week team baseline

Tender package

ClaudePythonParametric estimationFinancial analysis
automation · Active

Most business AI does not need to be a chatbot.

Tender Radar

A hands-off cron agent that reads new federal tenders, decides can-bid or cannot-bid against a company profile with the single decisive clause, and emits a weekly digest.

918 open federal notices at last count

Live feed measured

PythonClaudeCanadaBuys dataCron
content · Active

Retrieval over an owned corpus beats generation from model memory.

Scribe / Scribe Scout

A content engine that drafts from an atom graph of my own knowledge corpus, plus a discovery front end that scans the last 30 days of online discussion and scores topics by argument intensity.

Every draft built from my own atom graph, not model memory

Grounding

PythonClaudeReddit / HN dataKnowledge graph
healthcare · Pilot Complete

Real early-stage building, with the failure modes left in the write-up.

Healthcare Workflow Automation

Early-stage clinical workflow automation, piloted hands-on across three Nigerian clinics, cutting administrative documentation time by roughly a third.

Documentation time cut by roughly a third in pilot

Administrative time

PythonLangChainFastAPIPostgreSQL