I've used Gemini Deep Research for months, it is great to go deep into new topics and has helped me a lot. It has an issue though, ask for 5-10 researches focused on details and the majority will be wasted on the introduction and already explored topics. It wasn't able to build on top of previous work.
The solution came a few days ago when Obsidian CLI was released. The zettelkasten method, a way to organize your vault (that I learnt from this video https://www.youtube.com/watch?v=IMfz9E7g-Hk) provided a way to organize research atomizing knowledge and interlinking notes in an organized way. The skill uses Claude Code as the orchestrator and researcher, and Obsidian provides a way to structure the knowledge base in a way both agents and humans can interact with confortably.
The skill goes through this loop:
1. You feed it a topic or an existing Deep Research report.
2. It uses web search and extracts concepts into atomic markdown notes ([PN-...]).
3. It links them together under Maps of Content (MOCs) and checks the vault to avoid duplicating concepts.
4. It surfaces open questions and gaps. If you say "go deeper," it researches those specific gaps.
It is still a work in progress, but wanted to share it now because it has improved a lot my research workflow. I've been using it to research on how to do chinese web scraping and the results have been amazing. I'm worried about it growing without limits, maybe a pruning workflow can solve this?
p.d: massive thanks to all the zettelkasten comunity that made resources that helped me implement this.
I've used Gemini Deep Research for months, it is great to go deep into new topics and has helped me a lot. It has an issue though, ask for 5-10 researches focused on details and the majority will be wasted on the introduction and already explored topics. It wasn't able to build on top of previous work.
The solution came a few days ago when Obsidian CLI was released. The zettelkasten method, a way to organize your vault (that I learnt from this video https://www.youtube.com/watch?v=IMfz9E7g-Hk) provided a way to organize research atomizing knowledge and interlinking notes in an organized way. The skill uses Claude Code as the orchestrator and researcher, and Obsidian provides a way to structure the knowledge base in a way both agents and humans can interact with confortably.
The skill goes through this loop: 1. You feed it a topic or an existing Deep Research report. 2. It uses web search and extracts concepts into atomic markdown notes ([PN-...]). 3. It links them together under Maps of Content (MOCs) and checks the vault to avoid duplicating concepts. 4. It surfaces open questions and gaps. If you say "go deeper," it researches those specific gaps.
It is still a work in progress, but wanted to share it now because it has improved a lot my research workflow. I've been using it to research on how to do chinese web scraping and the results have been amazing. I'm worried about it growing without limits, maybe a pruning workflow can solve this?
p.d: massive thanks to all the zettelkasten comunity that made resources that helped me implement this.