Not a consultant with a deck. Not a vendor with a roadmap. C-suite AI leadership — on a fractional basis.
Three decades inside the institutions that define enterprise standards — and across the boardrooms of the banks and brands that set them.
Paul designed and shipped a production retrieval-augmented generation system — the same architecture pattern enterprise AI teams build. It runs live on this site.
Knowledge Base — Obsidian Vault. Roughly 2,000 interlinked markdown notes spanning eight books, original research, client frameworks, and a decade of writing. Plain text, version-controlled — the single source of truth the rest of the system reasons over.
Embedding Pipeline — Python · OpenAI. A Python job chunks the corpus and generates vector embeddings (OpenAI text-embedding-3), keeping the semantic index in sync as the vault grows. Built, run, and maintained by Paul — not a vendor.
Vector Store — Supabase · pgvector. Embeddings and metadata persisted in managed Postgres with the pgvector extension. Similarity search over the full corpus in milliseconds — the same architecture pattern production RAG systems are built on.
Orchestration — Claude API. A single Vercel serverless route (api/query.js) retrieves the most relevant chunks, assembles grounded context, and calls the Claude API for synthesis — answers anchored to Paul’s own writing, not the model’s priors.
Chat Interface — Public /corpus. The live, public Corpus page. Anyone can interrogate the body of work and get cited, grounded answers. A working demonstration — not a slide — of retrieval-augmented generation in production.
Eight books, two current whitepapers, four frameworks, and a working publication. The point of view is not borrowed.
Four domains of AI leadership, on a retainer. Click any domain to see what the work involves.
A short, direct conversation. No pitch deck. We will know quickly whether there is a fit.
Start the conversation →