Disclaimer: This project is an independent, non-commercial initiative and is not affiliated with, endorsed by, or associated with arXiv, Cornell University, Google, Gemini, Gemini CLI, or any related organizations, platforms, or technologies. All trademarks and names are the property of their respective owners.

Introduction

In this post, I want to document how I built an automated paper search engine using simple tools to help find, download, and summarize interesting academic papers in an automated way.

Nowadays, there are a number of tools and websites that help automate the process of finding academic papers. It would probably have been easiest to simply use those websites. But then this story would already be over and nothing would have been learned from it.

Vibecoding?

With the help of Gemini CLI, you can write (or have it create) wonderful programs. I was immediately drawn to the terminal-based user interface. But what should I do with it? What should I create?

Academic Search

Currently, ChatGPT can only summarize PDF files that you upload into the AI. However, this process is time-consuming and very limited unless you pay. Gemini CLI, on the other hand, can read and summarize multiple locally stored PDF files—but not going online and downloading papers and then summarizing them…

Building the academic search engine

One of the most wonderful sites for academic papers is still arXiv, in my opinion. The papers on this website come from fields like mathematics, physics, and computer science, and some of them are really complex. I love getting lost in these challenging texts…

So it became clear to me that I wanted to write a program that allows you to search for and download papers from arXiv via the terminal. Thanks to the API provided by arXiv, this was easy, and after a few hours of vibecoding, the CLI search for arXiv was born.

The two most important features are:

• Searching for papers on arXiv
• Downloading papers from arXiv

Adapted for Gemini CLI

Originally designed for human users, I had initially tailored the menu accordingly. However, I now wanted Gemini CLI to run the program itself—so I had to restructure it. This adaptation enabled the functionality of terminal commands—that is, direct execution in the terminal.

Read more about the command line search mode here

All md files

Instructions.md

With that in place, Gemini CLI was able to execute this Python program and thus find and download papers from arXiv. All it needed was an instruction and a list of topics to search for.

### Instructions for Gemini

#### Start arXiv Search Engine

1. Go to /folder/cli-search-for-arXiv-main (from now on declared as "arXiv Search Engine") read usage.md and then run the program run.py to search for articles.

#### Search for articles

2. Search for articles about the topics mentioned in topics.md 

Memory fact: ONLY use the program found in "arXiv Search Engine" to search for papers about these topics.

#### Download papers

2. Download these papers. How to do this you will read in "arXiv Search Engine" usage.md

#### Summaries

3. scan/read analyze these papers and write a summary.md file for each. 

4. Save the summaries in a folder where the name of the folder is the topic you searched for.

topics.md

### Topics

Complexity of universe
Complexity of chaos
Quantum Chaos and Universe
Quantum Chaos and complexity

Blocked? — Building my own PDFreader

Everything was working fine—until Gemini CLI suddenly blocked me, and it was no longer possible to read local PDF files. Don’t believe me? Here’s the proof.

That was frustrating—but I asked Gemini CLI for an alternative, and I got one. We simply built our own PDF reader and ran it locally. It was done in no time and worked beautifully! So we created the PDFreader and I gave Gemini the instruction to read summarize_instructions.md to summarize PDFs for me, by using the PDFreader.

summarize_instructions.md

### Instructions for Gemini

#### Run PDF reader

1. Go to /PDFreader and analyze this program. Understand it.

2. Run the program to read the PDF files in /folder

3. Summarize each of them and put the summary into a .md file with the same name as the PDF file. Use the format specified in summarize_format.md

4. Save the file in /DONE

summarize_format.md

### Formatting Instructions for Gemini

#### Summary format

1. Abstract / Summary

2. Key findings (bullet points)

3. Conclusion

4. Further reading

5. Important to consider

Conclusion and Findings

• Build small and simple programs that Gemini (or Claude) can execute to expand your AI’s capabilities
• Separate instruction files into multiple .md files to better organize and manage your prompts and instructions

Next steps

Now that you have all these .md files with summaries done, you can ask Gemini CLI to write a scientific paper for you, using all the information from these .md files. It does this wonderfully and while I would not publish the paper as it is, you can be at least be certain that the information in it is correct as it is taken from the summaries that were initially created.

How you use it is up to you…

Summary

Resources

📌 CLI search for arXiv (not affiliated with arXiv)

📌 PDFreader (python script)

📌 Gemini CLI

Prompts

You only need one prompt:

Please read and do instructions.md and then summarize using summarize_instructions.md and summarize_format.md.

…but beware as this might happen after a few minutes: