SaaS · 2025

CreativeNode.

Launching an AI-powered social media design SaaS from zero to working beta — generated posts, captions and hashtags for East African SMEs.

The brief

Where it started.

CreativeNode set out to build a SaaS that helps Kenyan SMEs create professional social media content without design skills. The product generates post designs, captions and hashtags using AI, scoped to the Kenyan market (Swahili long-tail, local context, KES pricing).

We partnered with the founding team on the entire zero-to-one journey — product design, front-end and back-end engineering, auth, billing, and the AI generation pipeline.

The challenge

What was hard.

AI-generated design is fast but often looks generic. The bigger design challenge was a template system that constrains the AI's output to on-brand visuals — colour tokens, type scales, component variants — so each SME's output looks cohesive, not stock.

Then there's cost control. OpenAI image generation is not cheap at scale, and an uncapped free tier would hemorrhage money. We had to build usage metering, credit packs, tiered pricing, and graceful degradation from premium to cheaper models as credits ran out.

Our approach

How we solved it.

  • 01
    Template-constrained generation
    Each brand has a template library — font pairs, colour tokens, component variants. Generation fills templates, not blank canvases.
  • 02
    Metered usage + credit packs
    Free tier (10 credits / month), paid packs (100, 500, 2000). Credits are atomic and transactional — billing never over-charges under concurrency.
  • 03
    Hot-swap model strategy
    Uses a higher-quality model for paid users, a cheaper one for free-tier. Both exposed through one internal API so the business logic doesn't know or care.
  • 04
    Swahili + English voice training
    Caption and hashtag generators trained on Kenyan-market phrasing. Supports English, Swahili and Sheng outputs.
  • 05
    M-Pesa subscription billing
    Daraja C2B for credit packs. Stripe for diaspora / USD subscribers. Both reconcile to the same credit ledger.
Stack

What we built it with.

Next.js 16React Server ComponentsPostgres + PrismaOpenAI / Anthropic APIsClerk (auth)M-Pesa Daraja + StripeVercel + Cloudflare R2PostHog analytics
What we shipped

The outcome.

  • End-to-end product from design through deployment.
  • Template-constrained AI generation on-brand per user.
  • Credit metering + tiered pricing with M-Pesa billing.
  • Auth, user dashboard, template editor, generation history.
  • Observability with PostHog + Sentry.
  • Production-ready admin for customer support.
2025Year shipped
KES/USDDual-currency billing
3Output languages (EN / SW / Sheng)
More work

Other things we shipped.

Chat on WhatsApp