Case study

Cadence

A content-automation platform that does everything up to the moment of publishing, then stops and asks a human. It grew a real media brand to 15,000 followers and monetized it. No AI slop.

RoleDesigned, built, and operate it solo
Year2024
StatusLive, monetized (15K followers)
Stackn8n · Redis · Flux LoRA · Telegram
The problem

Do it by hand, or hand it to a bot.

Running an active, high-quality social presence forces a bad choice. Do it by hand and it costs hours every day: research, write, source media, schedule. Hand it to a bot and it is fast, but generic and low quality, the "AI slop" that audiences tune out and platforms penalize.

Cadence is the third option: the speed and tirelessness of automation, with human judgment kept in the loop at the one place it matters, the decision to publish.

What I built

Twelve workflows, one editorial machine.

I designed and built the whole system end to end and operate it as the sole user: the multi-workflow architecture and orchestration, the ingestion connectors across five source types, the persona-driven rewriting engine, the deduplication and state layer that keeps it idempotent, a Notion-backed editorial queue, the Telegram human-review loop, and a custom-trained image model that gives the brand a signature identity.

It is not a side demo: it runs the brand's entire publishing operation, live since 2024.

How it works

Automation up to the publish, then you.

01 Ingest the sources →02 Deduplicate stories →03 Persona rewrite →04 Generate media →05 Propose the post →06 Human approval →07 Publish live

✓ you approve on Telegram before anything publishes. Refine the wording with the model, swap the media, or send it live.

Key features

What makes it more than a bot.

Multi-source ingestion

Reddit (API), news RSS, Telegram, Discord, and a personal Notion base, with configurable topics.

Persona voice engine

Hand-crafted prompts rewrite every item into one consistent brand voice.

Redis idempotency

A Redis-backed dedup and state store means nothing is processed or published twice. Safe to re-run.

Notion editorial queue

A status-driven state machine gives a hand-picked publishing lane alongside the automated one.

Signature image style

A custom-trained Flux LoRA gives the brand a recognizable visual identity.

Natural scheduling

Randomized post timing keeps the rhythm human, not robotic. Publishing is platform-agnostic.

Tech stack

Under the hood.

Orchestration
n8n, 12+ interconnected workflows
Ingestion
Reddit API · news RSS · Telegram · Discord · Notion
Language models
Self-hosted open-weight models (Llama 3, Mistral) via Ollama and HF Inference
Image generation
ComfyUI with a custom-trained Flux LoRA
State and dedup
Redis
Editorial queue
Notion (status-driven state machine)
Human interface
Telegram bot
Publishing
X (Twitter) API via OAuth, platform-agnostic by design
Outcome

Unattended for the work, human at the wheel for the call.

Built and operated for a personal media brand that Cadence helped grow to 15,000 followers and monetize. It runs unattended for the heavy lifting while keeping a human in control of every publish decision, which is what keeps the output quality high and on-brand.

Want automation that doesn't make slop?

I build automation that ships real output on a schedule, with a human in control of every publish.