Building in public · Not selling

LinkedIn Ghostwriting AI

An AI engine that writes LinkedIn posts in your voice. Not a ChatGPT wrapper. Custom RAG pipeline, tone cloning, and content strategy built in.

RAG Pipeline: Building Tone Cloning: Not started Scheduling: Not started Analytics: Not started

Why I'm building this

Ghostwriters are expensive

Good LinkedIn ghostwriters charge $500-3,000/month. Founders and indie hackers can't justify that early on. This engine bridges the gap.

ChatGPT output is generic

Most AI writing tools produce the same "In today's fast-paced world..." sludge. This one clones your actual tone from your past posts and learns your domain.

Consistency is the hard part

Writing one post is easy. Writing 3 per week for a year is not. The engine handles ideation, drafting, and scheduling so you just review and approve.

Real content strategy

Not random posts. A mix of authority builders, personal stories, hot takes, and curation — calibrated to your goals and audience.

Architecture

1. Content ingestion

Scrape or upload your past LinkedIn posts, articles, and tweets. Clean and embed into a vector store.

2. Tone extraction

Analyze writing patterns: sentence length, vocabulary, humor style, formatting habits. Build a tone profile.

3. Content strategy engine

Generate a 30-day content calendar based on your niche, goals, and audience. Suggest post types and angles.

4. Draft + review loop

AI drafts posts in your voice. You review, edit, approve. Feed edits back to improve future drafts.

Engine components

Vector store (ChromaDB)

Stores embeddings of your past content for retrieval. Queried during draft generation to pull in your actual opinions, stories, and vocabulary.

LLM router

Different models for different tasks: fast models for ideation, capable models for drafting, cheaper models for headline variants.

Content calendar

Drag-and-drop calendar with post type distribution. Ensures you're not posting the same type of content every day.

Feedback loop

Every edit you make feeds back into the system. After 20 posts, it should need minimal corrections. After 100, it should sound indistinguishable from you.

Roadmap

Done
Scraping pipeline, vector store setup, basic RAG
Doing
Tone extraction, draft generation loop
Next
Content calendar UI, scheduling integration
Later
Analytics dashboard, multi-platform support