jung-wan-kim

jung-wan-kim / autoresearch-builder

Public

Autonomous experiment loop for any project type (ML/Web/Flutter/Java). Inspired by karpathy/autoresearch, adapted for Claude Code.

14
3
100% credibility
Found Mar 20, 2026 at 14 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Shell
AI Summary

A plugin for coding environments that runs an infinite loop of suggesting file changes, testing them against a performance metric, keeping improvements via version control, discarding failures, and providing a dashboard for results.

How It Works

1
🔍 Discover the Helper

You learn about a clever assistant that automatically tests ideas to make your coding project better, inspired by a popular tool.

2
📥 Add to Your Space

You simply bring this assistant into your coding workspace so it's ready to use anytime.

3
📁 Choose Your File

Pick the main file in your project that you want improved, like your core script or app entry point.

4
🚀 Launch Auto-Improver

Start the endless loop where it dreams up changes, runs quick tests, keeps winners based on results, and tosses losers.

5
📊 Check Progress Dashboard

Open the neat dashboard to see charts of experiments, what worked, improvements tracked, and overall trends.

🎉 Celebrate Smarter Project

Your file is now optimized with better performance, shorter code, or faster runs, all handled automatically.

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 14 to 14 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is autoresearch-builder?

This shell-based plugin for Claude Code runs an autonomous experiment loop on your project: it analyzes the current state, tweaks a single target file, executes builds or tests, parses a key metric like bundle size or validation loss, and commits improvements via git only if better—discarding failures instantly. Inspired by karpathy/autoresearch but adapted for any stack from ML training to web apps, Flutter APKs, or Java builds, it auto-detects project types and logs everything to TSV and JSONL files for easy review. You get hands-off iteration via slash commands like `/autoresearch` or `/autoresearch results`, solving the grind of manual optimization.

Why is it gaining traction?

Unlike the original's ML-only focus and GPU demands, this handles web bundle sizes, APK footprints, build times, or custom metrics across stacks—no setup hassles, just drop a CLAUDE.md config. Developers dig the infinite loop that favors deletions and simplicity, plus git branches and a dashboard script for tracking trends in autonomous experiments. It's a natural fit for autonomous coder github workflows or autonomous exploration github, bridging to fields like autonomous driving simulation github or autonomous experimentation for accelerated materials discovery.

Who should use this?

ML engineers tuning models without GPU lock-in, Flutter devs chasing smaller APKs, Node.js teams shrinking bundles, or Java builders cutting compile times. Ideal for autonomous car project github tinkers, autonomous driving dataset github curators, or researchers in autonomous efficient experiment design for materials discovery with bayesian model averaging—anyone running repetitive experiments in Claude Code.

Verdict

Try it if you're in Claude Code and want autonomous github copilot-style tweaks; docs are solid and config is straightforward. With 14 stars and 1.0% credibility score, it's immature—low adoption signals risks like edge-case bugs—but promising for niche autonomous experimentation systems for materials development.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.