One Platform, Two Products: How We Build SaaS at Nukes AI

We run two in-house products. Repli triages cold-email replies; Cited tracks how often AI engines recommend your brand. On the surface they have nothing in common — different buyers, different problems, different data. Under the surface they share roughly two-thirds of their production foundation. That's not an accident; it's the strategy.

This post is about the platform-first approach that lets a small team ship more than one product without the second one taking as long as the first.

The part that's the same in every SaaS

Strip a B2B SaaS down and a large, unglamorous layer is identical regardless of what the product does: authentication and onboarding, subscription billing and metering, role-based access control, multi-tenant data isolation, an admin console, audit logging, a notification system, the integration framework for Slack and webhooks and Zapier, plus the observability and security spine that keeps it all honest.

Most of that is non-differentiating. No customer chooses Repli or Cited because of how its RBAC is implemented — but every customer leaves if it's done badly. It's exactly the kind of work you want to build once, harden under real load, and never rebuild.

Build the platform once, point it at two problems

So that's what we do. The platform layer — auth, billing, RBAC, security, observability, admin, notifications, integrations — is built and hardened as its own concern. Each product is the thin, differentiated layer on top: Repli's reply-ingestion pipeline and voice-cloning drafts; Cited's multi-engine scan scheduler and citation-extraction engine.

When we started Cited, we didn't re-solve authentication, billing, or alerting. We inherited a layer that had already survived contact with real users in Repli. That's the entire advantage of running both under one roof: one hardened platform, two revenue surfaces, and a second product that reaches production far faster than a from-scratch build.

The discipline that makes it work

Reuse only pays off if the shared layer is actually clean. A tangled platform shared across two products doesn't halve your work — it doubles your blast radius. So the boundaries matter: the platform exposes well-defined interfaces, each product depends on those interfaces rather than reaching into internals, and a change to billing can't quietly break citation tracking.

It's the same discipline we bring to client engagements — typed end-to-end, instrumented from day one, built to be operated rather than babysat. The difference is that our own products are where we get to prove it on our own time, under our own name, and carry the lessons straight back into the work we do for clients.

If you're building something that needs that kind of foundation — or you want to see the approach in practice — Repli and Cited are both in early access, and our studio takes on a small number of client builds. Either way, this is how we think about shipping software that has to actually work.