LinkedIn Content Intelligence Engine

Delivery approach

1. S

   Situation

   Finding a timely technical topic meant checking several communities, removing duplicate stories, judging audience fit, and turning the source material into an original post. The manual work took more than 20 hours each week. One-shot generation produced repetitive hooks, weak attribution, and a different voice from one run to the next.

2. T

   Task

   Build a local workflow that found current GenAI discussions, ranked useful ideas, and produced LinkedIn drafts for review without hiding the sources or fixing editorial choices in code. It needed fast and deep modes, visible progress, token and worker limits, and structured output for A/B testing. Posting had to remain a human decision.

3. A

   Action

   Source adapters collect posts from Reddit, Hacker News, Dev.to, and Stack Overflow and normalize their URLs, timestamps, engagement, authorship, and text. Canonical URL and semantic checks remove duplicate coverage. The remaining stories are scored for freshness, audience fit, technical substance, and novelty. LangGraph carries the selected evidence through angle selection, hook generation, drafting, sanitization, and review, so a later step cannot replace the premise without a source. Fast mode limits sources and passes; Deep mode allows more analysis within a set token budget. Worker limits protect source APIs and local resources, and retries retain completed stages. The JSON export contains the draft, alternate hooks, sources, selection rationale, and review findings. Server-sent events show progress in the browser. Regression examples test factual support, repeated phrasing, retained links, tone, and schema validity. The workflow ends at export for human review.

4. R

   Result

   Research and drafting time fell by more than 20 hours per week. The structured export supports A/B tests for hooks and angles, and every draft keeps links to the discussions behind it. Editors can check a weak claim before publication and change source, budget, or audience settings without editing the pipeline.