Case study · AI-native marketing operations
A content engine, not a content calendar.
Built, operated and used daily · Python, FastAPI, ChromaDB
Most marketing teams treat market intelligence and content production as two jobs. I built a system where the first one feeds the second automatically, and a human only decides what ships.
How it works
- Ingestion: around 30 curated specialist sources, newsletters via the Gmail API, RSS and YouTube transcripts, normalised and embedded into a local vector store.
- Intelligence: semantic clustering groups the same story arriving from different places, then each cluster is scored on how many independent sources carry it, how recent it is and how substantial the coverage is. What matters rises; noise stays down.
- Generation: ranked topics become multi-format drafts in a consistent voice: posts, infographics, video scripts, carousels. Visuals are generated alongside the copy, not requested afterwards.
- Publishing: everything lands in an editorial calendar as draft, review, approved, posted. Nothing publishes without a human.
Why this matters commercially
Competitive intelligence usually arrives as a quarterly document that is out of date the week it lands. Here it is a standing state of the market that also happens to produce the raw material for thought leadership.
The same pipeline answers “what is happening” and “what should we say about it”.
What transfers
- I build the tools I work with, which means I know from practice where these systems hold and where they quietly do not.
- Scale without losing voice is a design problem, not a volume problem. The approval step is the point, not the bottleneck.
Built and operated privately for my own market observation and publishing. Not a commercial product.
Questions about this? A short call is the quickest way. Book a call