A multi-tenant AI platform that drafts on-message, inside the lines.

Context

Organizations that live or die on message (campaigns, advocacy groups, trade associations) spend twenty to forty hours drafting a single position paper, and still risk going off-message or over a red line. Generic AI writing tools don’t know the organization’s positions and don’t know its limits.

What we built

We built a multi-tenant AI SaaS that generates strategic content grounded in each organization’s own knowledge base. Documents are ingested (PDF, Word, web) and indexed for hybrid vector-plus-keyword retrieval on pgvector, so every draft is retrieval-augmented against real source material. A three-tier boundary classifier (safe, caution, off-limits) keeps output inside declared positions. It produces six content types (position papers, talking points, briefings, press releases, speeches, op-eds) and is fully multi-tenant: per-client memberships and invitations, row-level security isolating each tenant’s data, and Stripe subscription billing with guest checkout.

The value

• Cuts a twenty-to-forty-hour position paper to a few hours, grounded in the organization’s own material rather than a model’s guesswork.
• The boundary classifier keeps AI output on-message and inside declared limits. The thing organizations fear most about AI writing, handled.
• Every tenant’s knowledge base and drafts are isolated by row-level security, real multi-tenant SaaS, not a shared bucket.
• A complete commercial product, not a prompt wrapper: ingestion, retrieval, generation, and billing.