Skip to content
Work

AI Content & Marketing

AI Social Media Publishing Platform

Architecture for a platform where users generate social content and images with AI agents, then preview, schedule, and publish across multiple platforms from one dashboard.

Next.jsConvexAI AgentsLate APIReplicateFLUXKimi APITypeScript

Overview

Designed the architecture for an AI-powered content generation and publishing platform that enables users to generate social media content through AI agents, create images, preview posts, and publish or schedule them across multiple platforms from a single interface.

Problem

Producing on-brand social content and getting it live across multiple platforms are two separate jobs — writing/design tools don't publish, and publishing tools don't generate — forcing teams to stitch several point solutions together by hand.

Business Goal

Give a user one interface to generate content and imagery with AI, preview exactly what will be posted, and publish or schedule it across platforms — collapsing the content and distribution steps into one workflow.

Architecture

System design

A Next.js dashboard backed by Convex handles auth, scheduling state, and user-provided API keys. AI agents generate copy variations per platform; Replicate/FLUX generates accompanying imagery; the Late API handles multi-platform publishing and scheduling through OAuth-connected accounts, with a preview layer showing exactly how each post will render before it goes live.

Challenges

What made this hard

Each social platform has different content constraints (length, image ratio, formatting) for the same underlying post.

Content generation is platform-aware from the start — the AI agent produces per-platform variants rather than one post force-fit into every format, with the preview layer catching mismatches before publishing.

Letting users bring their own API keys means the architecture can't assume a single shared provider account.

Designed a per-user credential layer so API keys and connected accounts are scoped to the user, keeping usage and cost attribution correct.

Business Impact

What it changed

Created a scalable architecture for managing AI-generated marketing content across multiple social networks from one centralized platform, rather than juggling separate generation and publishing tools.

Lessons Learned

  • Designing the preview layer early forced good decisions about platform-specific formatting that would have been painful to retrofit.
  • Treating scheduling and publishing as a first-class backend concern (not a UI afterthought) made multi-platform support far more tractable.

Future Improvements

  • Add analytics feedback so AI content generation can learn from what actually performs per platform.
  • Support team accounts with shared content calendars and approval steps.