I built WriteRAG to apply knowledge from professional books on writing to a specific text. The goal was feedback that explains which principle to consider, why a point matters and how the text could be improved.
How the review works
I collected writing principles, explanations and examples from books, keeping references to the sources. I used a vector database to retrieve relevant material.
When you submit a text, the system finds related principles. AI checks whether they apply and prepares feedback with explanations, recommendations and source references. The aim is to support both editing and learning from your own writing.
My contribution
I structured the knowledge base and built the workflow connecting retrieval, text analysis and editorial feedback.
The outcome
The system works and is in use, but the implementation proved slow and awkward to use. In practice, vector search did not meet my expectations.
My rough cost estimate for an in-depth review of one post is about $10, and in some cases around $20.
WriteRAG remains a useful experiment for me: detailed feedback is available and can support learning, but speed, usability and cost limit its practical use.