Entity Memory Database

PostgreSQL/Supabase project for canonical entities, generated dossiers, source provenance, activity timelines, commitments, evidence, RLS, and audit logs.

Overview

Built a database-backed version of an LLM-maintained knowledge wiki for business entities and operating context, inspired by Andrej Karpathy’s LLM Wiki pattern.

Instead of saving generated knowledge as loose markdown pages, this project used PostgreSQL/Supabase to store entities, generated dossier pages, source provenance, activity logs, commitments, and handoff records across prospect, client, and operations workflows.

What It Does

The project turns scattered business artifacts into records that can be queried, audited, and reused:

  • Canonical entity records for people, companies, prospects, clients, and related business objects
  • LLM-maintained profile and dossier records
  • Source-level provenance for activities and notes
  • Commitment tracking with evidence-based closure
  • Cross-schema handoff from pipeline context into client and operations workflows
  • Row-level security, audit triggers, and compatibility views for automation workflows

Technical Focus

  • PostgreSQL / Supabase schema design
  • PL/pgSQL triggers and migration-safe compatibility layers
  • Canonical entity resolution and backfills
  • Row-level security and audit logging
  • n8n and agent workflow integration
  • Activity and commitment models tied back to source evidence

Why It Matters

Handoffs often lose the useful details: discovery notes, entity facts, commitments, and evidence end up split across transcripts, calendars, forms, and automation logs. This project keeps those details in a structured database so agents and people can find the same source-backed record later.