Greg's projects hub for various legal AI tooling and personal automation

Agentic M&A due diligence pipeline — plan, classify, review in parallel, join the risks up, question the seller, draft the red flag report. Thirty-four skills, sixteen agents, thirteen workstreams.

Portfolio monitoring for private equity — covenant headroom, change-of-control consents, compliance boards and benchmark grids, every non-green cell cited to its clause.

Anonymises corporate transaction documents for a precedent database — entity discovery, preservation, Word tracked changes and an independent QA sweep, with three human checkpoints.

Standalone due diligence skills library — one jurisdiction-neutral core skill per workstream, with local-law overlay files layered on per jurisdiction. Tool-agnostic, no pipeline dependency, plus the teaching materials for authoring them.

Three entirely fictional M&A data rooms, each with a hidden answer key of planted findings — Project Halcyon (1,000 documents, twelve workstreams), its 3,000-document all-PDF second edition, and Project Argent, a 750-document carve-out room. The test bed behind the DD pipeline and the tool benchmarking harness.

Persistent memory across Claude Code, claude.ai and Cowork, backed by a Notion hub of projects, session logs and durable facts. Ships as a marketplace plugin from the gb-tools repo.

Six-feed open-data stack for a Surrey parish — planning applications, flood warnings, crime, power cuts, roadworks and a weekly digest — running in Docker on a Mac Mini, sinking to Notion.

n8n in Docker on the Mac Mini behind Tailscale, wired to Claude Code via MCP. First resident workflow polls Simon Willison's feed every three days and archives each post to Notion, with a heartbeat stamp so a stalled poller is visible.

Clears a five-thousand-item Readwise Reader backlog — dedupe, metadata triage, model classification into tag buckets, then an opt-in archive sweep. Dry-run by default; nothing is ever deleted.