SocialPilot — AI Social Media Automation Platform

Overview

SocialPilot is the largest thing I have built. It is a self-hosted social media marketing platform that takes a single campaign brief and turns it into finished, platform-specific posts — generating copy, images and video with AI, holding everything in an approval queue, then publishing to Instagram, Facebook, TikTok, YouTube and a website. I designed the product, wrote the code, deployed it, and use it as an operator on live client work.

Challenge

Running multi-channel content by hand does not scale. One campaign idea becomes a caption for Instagram, a different caption for Facebook, a vertical crop for TikTok, a thumbnail for YouTube, a scheduled slot for each, and a follow-up when a lead comes in. Doing that properly for one client fills a week. Doing it for several is impossible without either hiring people or building something.

Audience

Small and mid-sized businesses that need consistent social output but have no in-house marketing team — the exact profile of the clients I already work with: a construction company, a property agency, a detailing business.

Objectives

• Turn one brief into publish-ready content for every connected platform
• Keep a human approval step so nothing goes out unreviewed
• Generate original images and video rather than relying on stock
• Handle the whole loop — plan, create, approve, schedule, publish, capture leads
• Run it on infrastructure I control, at no per-seat cost

My responsibilities

• Designed the product: information architecture, workflow and interface
• Built the API in Fastify with Prisma and a Postgres database, plus Redis and BullMQ for background jobs
• Built the web app in React with Tailwind and Radix UI components
• Integrated the Anthropic API for copy generation, plus AI image and AI video connectors
• Wrote OAuth connectors for Facebook, Instagram, TikTok, YouTube, Google Drive, Google Sheets, Google Business Profile and Telegram
• Built video processing with ffmpeg for per-platform rendition and cropping
• Deployed with Docker Compose behind a Caddy reverse proxy with automatic HTTPS
• Hardened and administer the Ubuntu VPS it runs on, with uptime monitoring
• Prepared the Meta App Review submission pack for Advanced Access

Research & insight

• Every feature came from a task I was already doing manually for a client — the product is a record of where the real friction was
• The bottleneck was never having ideas, it was reformatting one idea into five deliverables
• Clients will not accept fully automatic posting, so an approval queue is not optional — trust is the product requirement, not a nice-to-have
• Feed aesthetics matter to clients as much as individual posts, which is why planning tools beat scheduling tools alone

Strategy

• One brief in, many platform-native outputs out — never the same text pasted everywhere
• AI drafts, human approves: the owner keeps editorial control at all times
• Self-host on owned infrastructure so there is no per-seat SaaS cost as clients are added
• Build only what I had personally hit as a problem, so no speculative features

Channels used

Instagram

Publishing posts, reels and carousels via the Meta Graph API

Facebook

Page publishing through the same Meta connection

TikTok

Vertical video publishing connector

YouTube

Video upload connector

Website

Webhook publishing to a connected site

Creative execution

• Dashboard with autopilot toggle, connected platforms, pending approvals, engagement and lead counters
• Content Calendar with an AI composer — brief, optional template and asset, or generate a new AI image or AI video, then pick platforms and generate
• Campaign planner for grouping posts under a single campaign
• Reel Builder for assembling short-form video
• Grid Planner for laying out the Instagram feed visually before publishing
• Brand Workspace holding tone, colours and reusable templates so output stays on-brand
• Leads and Lead Magnets modules for capturing and following up enquiries

Deliverables

• Live self-hosted platform at marketing.elysian-estates.eu
• Fastify + Prisma API with background job queue
• React web application
• Nine third-party platform connectors
• Docker Compose deployment with automatic HTTPS
• Meta App Review documentation pack

Results

• Running live in production and managing a real client workspace
• Instagram and Google Drive connected and authenticated through OAuth
• AI copy, image and video generation working end to end inside the composer
• Meta App Review pack written and ready for submission to unlock client self-onboarding

Lessons learned

• Building the tool taught me the marketing work better than doing it manually did — you cannot automate a process you do not fully understand, so every feature forced me to define exactly what I was doing and why
• The approval step is the feature clients care about most; full automation sounds impressive but nobody wants to hand over their brand voice unsupervised
• Platform APIs are the hard part, not the AI — Meta's permission model and review process shaped the roadmap more than any technical decision

Media gallery