Build Advanced and optimized Research Workflows with n8n

more details about how to make your n8n better

Following the completion of my DeepMed AI project and launch on Product Hunt, I want to show you more details how I restructured the original Deep Research workflow shared by Jim_Le (see first article about it here) in the n8n community. My goal was to build a more domain-specific automation system tailored for biomedical research – and make it stable enough to support real-world academic or industrial use.

The original template I found on the n8n community forum used:

• Tavily for web search

• Gemini for summarization

• A simple output format

It worked great as a proof of concept – but for rigorous scientific workflows, it had a few limitations:

1. Search results weren’t always focused on biomedical sources

2. The AI’s output format was too casual for academic use

3. Prompts were generic and lacked depth

So I began restructuring the entire workflow.

Redesign Strategy: Scientific, Stable, and Scalable

Restricting Search to PubMed

To avoid pulling results from blogs or random forums, I added a domain filter in the Tavily search node, and this significantly improved the quality and reliability of the data being fed into the LLM.

domain: pubmed.ncbi.nlm.nih.gov

Providing a Clear Report Template

I created a structured prompt that asks the AI to return its output in the following format, which gives me clean, predictable output — perfect for turning into formal biomedical reports or integrating into other systems.

• Title
• Abstract
• Introduction
• Materials and Methods
• (Published) Results
• Underlying mechanisms
• Drug delivery system
• (Pre-) clinical models
• Conclusion and Discussion
• References (with PMID)

Using Follow-Up Prompts for Depth

n8n’s modular design allowed me to chain prompts. Here’s how it works:

• The user enters a topic (e.g., “siRNA in cancer therapy”)
• Gemini provides an initial overview
• Then a follow-up prompt is triggered: “Based on this, list three unresolved questions and explain why they’re important.”

This created a multi-layered reasoning process that mimicked real academic exploration.

Saving Output

All results were saved to Google Sheets with clearly defined fields – including topic, summary, open questions, and PMIDs.

Final Thoughts: Build Your Own Research System

What I ended up with wasn’t just a template – it became the foundation for a reusable, extensible research automation system. With this foundation, you can easily expand it to include:

• PDF extraction
• Google Scholar scraping
• Diagram generation
• Email summaries
• Presentation slide creation

n8n isn’t just a tool – it’s a flexible platform that empowers subject matter experts like us to define our own research assistants. If you’ve already experimented with basic n8n workflows, now’s the time to level up. And of course, if you get stuck or want to build something similar, feel free to reach out. I’d love to see what you create.