Built AI-powered research workflows with n8n — what used to take weeks or months, now takes under 10 minutes.
I’ve been experimenting with building workflows to automate country-level research. The system can now collect 40+ data points per country, pulling from multiple sources and formats.
Here’s what the workflows can do:
Call APIs from sources like World Bank and Our World in Data
Upload PDFs and auto-summarize the content
Process CSVs to fetch and format key data
For non-API sources: scrape web pages via HTTP requests + CSS selectors
Use AI prompts to generate research summaries from the collected data
Push everything into a central database for analysis
Some key takeaways from this build:
- Prompt engineering still matters – You need to guide the AI with the right questions and credible sources.
- Human input is essential – Local context checks are still important for accuracy.
- Documentation is non-negotiable – As workflows grow, clearly documenting each node’s logic saves hours down the road.
Had a lot of fun experimenting with the logic, and I’m excited about how much more scalable research can be with the right workflows in place.