This is a submission for the Gemma 4 Challenge: Build with Gemma 4
What I Built
I built Open-Rosalind, a Gemma 4-powered biomedical AI agent for reproducible life science research. It helps users ask biology questions through a simple web chat interface, while the backend routes each query to biological skills such as sequence analysis, protein annotation, literature search, and mutation assessment.
The main problem I wanted to solve is that general-purpose AI agents can sound confident, but biomedical research needs more than fluent answers. Scientific answers should be connected to tools, databases, papers, and reproducible calculations. Open-Rosalind is designed around a tool-first workflow: tools produce evidence, and Gemma 4 summarizes that evidence into a readable answer.
The long-term goal is to make biomedical agents more trustworthy and practical for real research environments, including local-first deployment with private sequence libraries, literature collections, and institutional biological data.
Demo
Live project:
https://openrosalind.bio
Video walkthrough:
Code
The project is open source here:
https://github.com/maris205/open-rosalind
The entire system was developed with the help of Codex and Claude Code.
How I Used Gemma 4
I used Gemma 4 27B MoE via OpenRouter as the core reasoning and summarization model behind Open-Rosalind. Gemma 4 helps the system understand user intent, route biomedical questions to the right workflow, and turn structured tool outputs into clear natural-language answers.
I chose Gemma 4 27B MoE because it offers a good balance of capability, efficiency, and accessibility through OpenRouter, while also fitting the long-term local-first direction of the project. In Open-Rosalind, Gemma 4 is not used as a standalone chatbot; it reasons over evidence produced by biological tools, so the final answer is more traceable and suitable for biomedical research workflows.
Top comments (0)