ScienceBuddy: Recursive-in-Recursive Self-Improvement for Scientific Agents
Shuhan Xue · hf · 2026-09-16
ScienceBuddy is an interactive research workspace that embeds continually improving scientific agents into researchers' workflows, converting requests, feedback, and execution evidence into tasks and eval rubrics. Its core is a recursive-in-recursive paradigm: the inner recursion evolves the harness with the model fixed, while the outer recursion trains the model under the improved harness. Case studies span four scientific task families; released at science-buddy.io.
More from coding & agent
- Workshop 0.3.1: Open-Source Execution Architecture for Grounded AI Work — Promptmethus · 2026-09-16
- Energon: Open-source self-hosted storage and publishing service built for AI agents — kieranklaassen · 2026-09-16
- lateinteraction: let agents compile expensive LLM calls into cheap AI functions, likely via DSPy — lateinteraction · 2026-09-16
- Dev builds MCP server feeding Claude curated UI/UX patterns, seeks feedback — Separate-Topic4883 · 2026-09-16
- SwiftUI co-creator defends client-side animations: dropping Core Animation was deliberate — JordanMorgan10 · 2026-09-16
- Building a browser game inside a Gaussian splat made in World Labs Marble — theworldlabs · 2026-09-16