liviaellen

liviaellen / autoresearch-gen

Public

Scaffold generator for Karpathy's autoresearch. PT + MLX, any LLM, Streamlit dashboard.

52
6
100% credibility
Found Mar 14, 2026 at 47 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

A generator that creates verified experiment setups for autonomous machine learning research, including data preparation, training scripts, agent instructions, and a visualization dashboard.

How It Works

1
🔍 Discover autoresearch-gen

You stumble upon this handy tool on GitHub that helps you set up and run machine learning experiments automatically overnight.

2
🛠️ Get ready quickly

You grab the simple setup tool called uv and prepare everything with a few easy steps—no complicated installs needed.

3
💭 Describe your idea

You share what you're researching, like your project details, the data you'll use, and what you hope to improve.

4
🗣️ Chat with AI helper

A friendly AI reads your notes, spots any gaps, and asks a few smart questions to make your setup spot-on.

5
✨ Build and test setup

It creates your own experiment folder, writes the starting files, and automatically runs a test to confirm everything works perfectly.

6
📊 Launch results viewer

You open a colorful dashboard that shows your verified starting results, charts, and insights at a glance.

🚀 Watch AI improve models

You hand it off to an AI agent that runs experiments all night, and you wake up to better-performing models and progress graphs.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 47 to 52 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-gen?

autoresearch-gen is a Python scaffold generator that automates setup for Karpathy's autoresearch, creating ready-to-run ML experiment directories with PyTorch or MLX backends. Describe your project context, data, and goals via CLI flags or interactive prompts, and it generates data prep scripts, training loops, and agent instructions—then auto-runs a baseline to verify everything works before handing off to any LLM agent. A Streamlit dashboard visualizes results across experiments, auto-detecting metrics like val_bpb.

Why is it gaining traction?

Unlike basic scaffold generators like django-scaffold-generator or rails custom scaffold generator, it uses an LLM-powered interview to clarify vague inputs and auto-fixes broken code via retries, ensuring zero manual debugging. Supports switching LLMs (Claude, GPT, DeepSeek) and backends seamlessly, with Makefile targets for gen, dashboard, and diagrams—perfect for overnight autonomous runs. The end-to-end verification and multi-experiment dashboard hook devs tired of copy-pasting train scripts.

Who should use this?

ML engineers exploring architectures or hyperparameters on Apple Silicon (MLX) or CUDA (PyTorch), especially those running LLM agents for autoresearch loops. Ideal for solo researchers testing ideas like attention-free models or LR ablations on datasets like TinyStories or FineWeb, without setup hassle.

Verdict

Grab it if you're into autoresearch—solid for quick, verified scaffolds and dashboard tracking, despite low maturity (18 stars, 1.0% credibility). Early stage means watch for roadmap features like persistent memory; test on a sample experiment first.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.