saadmsft

saadmsft / nanoresearch

Public

Tri-level co-evolving multi-agent research automation — a faithful re-implementation of arXiv:2605.10813 with a ChatGPT-style web UI.

saadmsft.github.ionanoresearch
10
1
89% credibility
Found May 18, 2026 at 10 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

NanoResearch is an autonomous research assistant that transforms a simple research idea into a complete, peer-reviewed paper. You start by telling it your name, field, and preferences. Then you describe any research question — from ecology to medicine to computer science. The system searches academic literature, generates and evaluates multiple hypotheses, designs a detailed experiment plan (with an internal reviewer catching any flaws), writes and runs the experiment code automatically, analyzes the results, and produces a finished LaTeX paper compiled to PDF. Throughout the process, you can pause to give feedback that shapes the direction. Over time, the system learns your personal preferences and improves.

How It Works

1
👋 You introduce yourself

You tell the assistant your name, field of research, and how you like to work — whether you prefer careful experiments or bold explorations.

2
💡 You share a research idea

You describe any research question you want to explore — from the impact of urban green roofs on birds to new drug delivery methods.

3
🔍 The assistant reads the literature

The system searches through millions of academic papers, finds the most relevant ones, and extracts key findings and numbers to ground your idea.

4
🎯 You pick the best direction

The assistant proposes several possible hypotheses ranked by how novel and exciting they are. You choose which one to pursue.

5
📐 A detailed experiment plan appears

The system drafts a complete blueprint for your experiment — what data to use, what to compare against, what metrics matter — and an internal reviewer checks it for flaws.

6
You can pause to give feedback
✏️
You refine the direction

You give specific feedback like 'drop the field-survey arm and focus on EEG data' and the plan adjusts accordingly.

▶️
You let it continue

If the plan looks good, you say so and the assistant moves ahead automatically.

7
🧪 Code writes and runs itself

The system writes Python code to run your experiment, tests it, and fixes any errors automatically — all in a safe, isolated environment.

📄 Your paper is ready

A complete research paper materializes section by section in LaTeX format, reviewed for quality, and compiled into a PDF you can download.

Sign up to see the full architecture

6 more

Sign Up Free

Star Growth

See how this repo grew from 10 to 10 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 nanoresearch?

NanoResearch is an autonomous research pipeline that transforms a research topic into a full academic paper. It runs a three-stage process: first it surveys literature and generates novel hypotheses, then it designs and executes small experiments in a sandbox, and finally it writes and compiles a LaTeX paper complete with internal peer review. The system is built in Python with a web-based chat interface that resembles talking to a research assistant. Each stage uses specialized agents powered by Azure AI Foundry, with a local language model that can be personalized to individual researchers using a technique called SDPO.

Why is it gaining traction?

The hook is the fully automated pipeline -- you give it a research topic, and it produces a draft paper. Researchers spend enormous time on literature surveys, experiment design, and writing; this attempts to automate all three. The multi-agent architecture means each stage has a dedicated role with tailored prompts, producing more coherent output than a generic chatbot. The learning mechanism is the real differentiator: when you correct the system, it fine-tunes a personal adapter via SDPO, so subsequent runs reflect your preferences.

Who should use this?

PhD students or researchers drowning in literature review and experiment logistics will see the most value. Machine learning practitioners prototyping research ideas quickly could use it to generate baselines and paper drafts. Individual developers exploring AI agent pipelines for academic automation will find a well-structured reference implementation. It is not yet suitable for production research workflows -- the project is at version 0.1.0 with limited documentation and an early-stage codebase.

Verdict

NanoResearch is an ambitious, well-engineered implementation of an automated research pipeline, but with 10 stars and the credibility score at roughly 0.9 percent, it remains a bleeding-edge project for experimental use. If you want to explore multi-agent research automation or experiment with personalized AI planners, this is worth a look. Treat it as a research prototype, not a production tool -- and expect to read the code and configuration to get it running.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.