LinkedIn Post GeneratorTrend-aware AI writing pipeline for LinkedIn.
01
What sparked it
I wanted real hands-on time with APIs and a retrieval based system I could iterate on. While adding connections on LinkedIn and reading newsletters, I kept wondering whether a pipeline could pull in trending material for me and draft posts in different voices, instead of leaving every post as a blank page problem. That curiosity became the prompt for this project.
02
The pipeline
Four layers in order. Ingestion pulls 75 plus daily items across Hacker News, Reddit, RSS feeds, and News API. Deduplication and relevance scoring uses multi-weighted similarity to collapse near duplicates and drop noise, cutting irrelevant items by 45 percent. Tagging runs a 78 entity classifier so posts can be filtered by topic. Drafting is a four stage OpenAI prompt stack that turns a surfaced item into a LinkedIn ready draft with tone controls.
03
Infrastructure
Vercel hosts the Next.js front end, Render runs the Python ingestion and pipeline jobs, and Supabase stores the normalized trend items and user drafts. Keeping these separated meant I could iterate on prompts without touching the ingestion loop.