Work · Case study

Social media without the calendar juggling.

A multi-tenant social media management platform for one operator running several brands. Three content generation modes, backed by an AI quality spine and eight publishing adapters, so research, drafting, scheduling, and analytics all live in one cockpit.

Live, in production ● Multi-tenant SaaS ● React + Express + PostgreSQL ● MCP server (13 tools) ● Meta, Google, TikTok, more
What it does

The whole content pipeline, from research to publishing, in one cockpit.

S0cial Master takes the operator through the entire content workflow: research, drafting, review, and publishing across nine networks. Each step has the operator's preferred level of AI involvement, so the workflow stays in one place instead of being stitched across five browser tabs.

The result is a single cockpit where one person can run the social presence of several brands across Meta, Google, TikTok, and LinkedIn, with the AI doing the research and the drafts and the operator staying in control of what goes out.

S0cial Master marketing landing hero
Efficiency & Accuracy

The same build approach for any complex AI system.

Orchestration

Hermes powered workflow

S0cial Master's development and operational pipeline runs on Hermes. The agent system handles goal decomposition, content generation, scheduled publishing, and the auditing pipeline that catches quality issues before a post ships.

AI Strategy

Right-sized model for the task

With three content generation modes, S0cial Master routes the right model to each task. Open-source and open-weight models handle research and drafting volume. Frontier models take the high-stakes work: brand-voice consistency checks, banned-phrase matching, the final quality pass before publish.

Deployment

Rigorous quality assurance

Before any content is published, S0cial Master employs frontier models for a comprehensive quality spine audit, validating hook scores, checking for banned phrases, and ensuring brand alignment across eight publishing adapters, guaranteeing high-quality output.

S0cial Master Autopilot interface

The Autopilot interface. Chat-style AI that pulls research, generates ideas, expands to drafts, and schedules across networks.

The AI quality spine

Every draft passes through the same checks.

Before any generated draft reaches the UI, it goes through the same quality layer: hook scoring, banned-phrase match, length sanity, CTA presence, and hook uniqueness against the last thirty hooks for that brand. If the hook scores below threshold, the system auto-regenerates up to a configurable number of retries.

The same spine serves Manual, Monthly Meeting, and Autopilot. The operator only sees posts that already pass the quality bar.

The publishing pipeline

One chokepoint. Eight adapters.

The publish-orchestrator is a single chokepoint for cross-platform publishing. Per-platform adapters translate each network's specific publishing requirements. Validation runs against each platform's rules. Quota checking happens before the post leaves the system. Delivery mode is derived from the brand's configuration.

Posts only reach the operator once they have actually shipped to the network.

S0cial Master Studio interface showing post kanban

The Studio. Five-column kanban: Suggestions, Approved, Defining, Ready, Posted. Every card has a quality flag and a publish state.

S0cial Master calendar view of scheduled posts
The stack

One operator's tool. The whole pipeline.

Runtime Node 20+, TypeScript 5.6 strict mode, tsx for dev.
Server Express 5, Drizzle ORM, multer, ws. Routes organised by feature (content, autopilot, brands, analytics, etc.)
Client React 18, Vite 7, Tailwind 3, Radix UI, TanStack Query, Wouter hash routing. Dev port 5001 in middleware mode (Express + Vite in one process).
Database SQLite locally, PostgreSQL when DATABASE_URL is set. Dual-mode schema bootstrap.
LLM providers Perplexity (default), OpenRouter, MiniMax on Anthropic-compatible endpoint. Per brand, per task.
Image generation MiniMax image-01 (default) or OpenRouter. Cloudinary for hosting.
MCP layer 13 tools, low-level stdio transport. Lets external agents (Hermes, Claude) operate S0cial Master as a service.
Deployment Railway, Procfile, health check at /api/health.
S0cial Master MCP integration showing tool list

S0cial Master as an MCP server. 13 tools, callable from any compliant agent.

The MCP layer

The operator can also be another AI.

S0cial Master exposes itself as a Model Context Protocol server. The 13 tools cover the full surface: pull research, generate drafts, schedule posts, check status, manage the calendar. Any MCP-compliant agent can drive it.

That means the operator is not the only one who can run the platform. A scheduled agent can pull weekly research, generate a Monthly Meeting strategy, and surface it for approval. The operator stays in the loop on quality. The calendar is no longer the operator's job.

One operator, several brands, nine networks, eight publishing adapters. The calendar runs itself. The quality spine catches what would have slipped through.

What stands out

Single-tenant-per-brand on the inside, multi-tenant on the outside.

Every brand gets its own context: voice, banned phrases, top performers, last thirty hooks, the audience profile, the approved templates. The publish orchestrator knows which platform, which format, which schedule, which quota. The AI quality spine is calibrated to what 'good' means for that specific brand, its voice and its audience.

Multi-tenancy in the data layer, single-tenant in the experience. The operator of one brand never sees another brand's content, voice, or analytics. The platform operator sees the aggregated queue.

S0cial Master inbox view with message threads, sentiment flags, and quick-reply actions for each brand.
The bit underneath

Built on the same architecture I'd bring to yours.

S0cial Master is the case study. The build that produced it is the product. Hermes for orchestration. Open-source and open-weight models for the heavy lifting (research, drafting, image generation). Frontier models for the moments that matter (security audit, repo review, the high-stakes ten percent).

The MCP layer that drives S0cial Master from outside is the same MCP layer I use to build the platform. Same tool, different operator.

If any of this looks like a workflow you have, twenty minutes is enough to see if I can help.

Tell me which workflow looks closest to a problem you're sitting with right now. I'll tell you honestly whether AI is worth building it, and what the first useful step looks like.

Book a chat