The problem
A company's positioning, audiences, offers, evidence, and writing rules often live across documents and individual conversations. Reintroducing that context to every AI tool makes it difficult to keep outputs consistent and knowledge current.
My role
I lead the product and engineering work through Artebello, connecting knowledge structure, application workflows, AI integrations, and review behavior. Narravo is a live, actively evolving product; this profile describes implemented engineering capabilities, not an independently audited adoption claim.
The system
The Next.js application uses Supabase authentication and PostgreSQL persistence through Prisma. Its Living Brand Graph organizes company knowledge with sources, relationships, provenance, and review states. Approved context informs strategy and content creation.
A company-scoped MCP endpoint exposes context, search, and source retrieval. Connections have bounded permissions; knowledge suggestions enter a review inbox rather than becoming approved facts immediately.
Important decisions
- Keep candidate knowledge distinct from verified company knowledge.
- Apply company and source access boundaries before returning context.
- Separate portable business memory from a particular model provider.
- Make knowledge changes reviewable and attributable.
What the work demonstrates
The repository documents the product loop, knowledge export, private AI chat, and MCP implementation. Recorded engineering checks exercised protocol behavior, authentication errors, revocation, read-only permissions, and source isolation. These are implementation checks, not business impact metrics.
For the reasoning behind the integration, read connecting company knowledge through MCP.