I've been working on an open-source project called Deep Research OSS and would love to get some feedback from the community.
I was frustrated by AI research tools that only performed a single/double-pass search and often missed important details.
To solve this problem, I built a multi-agent pipeline that mimics how a human team works.
First, it creates a broad plan. Then, it uses agents to conduct research. Finally, it has another agent perform a gap analysis to identify missing information and assign new tasks to dig deeper. Additionally, to avoid errors, AI is forced to use code execution for all numbers in the report.
The whole thing is built specifically for the Gemini API to see how far it can be pushed. I've been using it for my own stock analysis, and the iterative process seems to generate much more comprehensive reports.
It's still a work in progress, and I'm sure there are better ways to approach this. I'm looking for contributors and feedback to improve the pipeline and make it more robust.
If you're interested in multi-agent systems or just want to help build a better research tool, you can check out the repo.