SEO description generation from a brief
Product descriptions were written by hand from a template and still came out inconsistent. The pipeline generates them from the client's brief and holds one tone of voice.
Outcome
- A description takes seconds instead of 15 minutes
- One consistent tone and structure across the catalogue
- Role
- Pipeline development
- Time it took
- 1 week
The problem
Product descriptions were written by hand from a template. The issue was not speed but consistency: different people on different days wrote differently, and the catalogue read like it had been stitched together from several shops.
What I built
A pipeline that generates a description from the client's brief: tone, required blocks, banned phrasings, keywords. The brief is input data, not a prompt, so the client can change the rules themselves.
Why the brief matters more than the model
There is one temptation here: ask the model to «write a good description» and take whatever comes out. That gives you text which looks fine as a single example and falls apart across hundreds. So the brief is strict: fixed structure, capped length, and the list of banned phrasings matters just as much as the list of wanted ones.
A check before it ships
Every finished description is verified: are all required blocks present, is the length within limits, did any banned wording slip in. If it fails, it is regenerated. That is cheaper than proofreading the whole catalogue afterwards.
Stack
- Python
- LLM API
- SEO
Got a similar task?
Describe it in two sentences. I'll tell you if it can be automated and roughly what it costs.