End-to-End Research with AI

Labour-market-AI-exposure case — question to paper, one checkpoint at a time.

1. setup 2. dataset 3. literature 4. hypothesis 5. design 6. analysis 7. results 8. manuscript 9. referee 10. disclosure 11. done

← Back to the course flow & decision graphs

1 · Set up your session

Your API key is kept only in server memory for this session (never written to disk or logs) and is discarded when the session expires or you close it. Nothing is shared with other visitors.

A small JSON file with any of api_key, semantic_scholar_key, scite_key, huggingface_token, kaggle_username, kaggle_key — read entirely in your browser to fill in the fields below. The file itself is never uploaded anywhere; only what you submit with "Start" is, exactly as if you'd typed it in by hand.
Recommended, not just optional — the public API rate-limits unauthenticated requests aggressively in practice, often on the very first search. A free key removes that. Get one at semanticscholar.org/product/api.
Paid (scite.ai) — if set, literature results also show whether later work supports or contradicts each citation.
Only needed to fetch gated HuggingFace datasets in the dataset-search step.
Only needed to fetch Kaggle datasets in the dataset-search step.
Default is the highest-quality choice. Switch to the Haiku-tier model for a cheaper, faster run.
You'll pick or generate a dataset for this on the next step. Answered as asked — the pipeline won't quietly substitute a different framing.
Shapes the manuscript's structure, tone, and length. Causal-language discipline and data/AI-use disclosure apply regardless of type.