# Team Darwin

- **Event:** [The Harness Engineering & Model Wrangling Hackathon](https://cerebralvalley.ai/e/mongodb-nyc-hackathon)
- **When:** Sat, Sep 26 at 9:00 AM – 10:00 PM (EDT)
- **Where:** New York, NY
- **Team:** [Iveena Mukherjee](https://cerebralvalley.ai/u/blueballoons25), [Vedic Mukherjee](https://cerebralvalley.ai/u/vedic)
- **GitHub:** https://github.com/VedicAM/darwin
- **Demo video:** https://drive.google.com/drive/folders/1XnfuMFlp2pcVgxVZgBwdMCxIW7CZDKZD?usp=sharing
- **Gallery:** https://cerebralvalley.ai/e/mongodb-nyc-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/mongodb-nyc-hackathon/hackathon/gallery/79

Darwin. Self-improving research system for 5S rRNA biology.

This project builds a self-improving research system to investigate how changes in 5S ribosomal RNA and its assembly partners, RPL5 and RPL11, affect ribosome production and cellular stress signaling. A motivating application is Diamond–Blackfan anemia, a rare disorder in which impaired ribosome production disrupts red-cell development. The 5S ribonucleoprotein complex offers a molecular bridge between RNA structure, assembly of the 60S ribosomal subunit, and the MDM2–p53 stress response.

The agent combines literature, sequence, RNA structure, protein structure, genetics, and expression evidence to generate **specific, experimentally testable hypotheses**. STAIR and ToC index available tools so the agent can select relevant capabilities efficiently; ToolUniverse provides a logged `find_tools` and `execute_tool` interface. Local analysis tools align sequences, model RNA folding, and compare results across datasets. Each conclusion retains its source and uncertainty.

The central research contribution is **recursive self-improvement of the research workflow**. The agent identifies failures in retrieval, tool selection, analysis, or verification; proposes a new cited skill or a change to its procedure; and keeps that change only if it improves performance on fixed, held-out tests. Historical publication cutoffs allow us to ask whether the system can anticipate later experimental findings without seeing them.

The intended result is a reproducible system that improves both the **quality of biological hypotheses** and the **process used to generate and check them**. Its first application would prioritize 5S RNP mechanisms for laboratory follow-up, rather than make clinical diagnoses from predicted RNA structure.

## More from The Harness Engineering & Model Wrangling Hackathon

- [Refresh](https://cerebralvalley.ai/e/mongodb-nyc-hackathon/hackathon/gallery/76)
- [Rulebook](https://cerebralvalley.ai/e/mongodb-nyc-hackathon/hackathon/gallery/77)
- [AI Cybersquad of 1](https://cerebralvalley.ai/e/mongodb-nyc-hackathon/hackathon/gallery/78)
- [Raaya](https://cerebralvalley.ai/e/mongodb-nyc-hackathon/hackathon/gallery/80)
- [Tomok](https://cerebralvalley.ai/e/mongodb-nyc-hackathon/hackathon/gallery/81)
- [JITAgent](https://cerebralvalley.ai/e/mongodb-nyc-hackathon/hackathon/gallery/82)

---

Markdown version of https://cerebralvalley.ai/e/mongodb-nyc-hackathon/hackathon/gallery/79. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt