Research Agent

An agent that retrieves, screens, and summarizes research papers using OpenAI and LangChain.
Explore the Docs »

Report Bug · Request Feature

Table of Contents

1. About The Project
2. Features
3. Built With
4. Getting Started
5. Usage
6. Roadmap
7. Contributing
8. License
9. Contact

🚀 About The Project

Summarizing scientific literature is time-consuming. Research Agent automates this process:

• 🔍 Retrieves papers from PubMed Central
• 📄 Screens abstracts using your review objective
• 📊 Performs structured meta-analysis or systematic review
• 💾 Outputs clean JSON with summaries and metadata

(back to top)

🛠️ Built With

• LangChain
• OpenAI GPT-4o
• Typer CLI
• Rich
• Python 3.10+

(back to top)

🧑‍💻 Getting Started

Prerequisites

• Python 3.10 or higher
• An OpenAI API key Get the key!
• A NCBI API key Get the key!

Installation

git clone https://github.com/MakanFar/research-agent.git
cd research-agent
pip install .

Configuration

Edit or create config.yaml in the root directory:

🔑 API Keys

• you can add your API keys to the config:

OPENAI_API_KEY: sk-...
NCBI_API_KEY: ...

Alternatively, you can export them as environment variables:

export OPENAI_API_KEY=sk-...
export NCBI_API_KEY=...

🔍 Default Search Query

Specify a PubMed Central query:

search_query: ("Artificial intelligence"[Title/Abstract] AND "Protein-protein interaction")

You can also provide the query directly via the CLI at runtime.

🧾 Metadata Extraction (from abstracts)

Customize fields to extract during meta-analysis or systematic review:

meta_data:
     - data_type: Type of data used in the study such as radiology, clinicopathologic, or text
    - species_breed: Target species
    - ml_algorithm: Types Model used in the study
    - ai_goal: Clinical objective of the study
    - performance_results: Key final performance results

Note: The agent always extracts the following by default:

• Year
• First Author
• Title
• Journal
• PMCID
• URL

📘 Full-Text Extraction (systematic review only)

Optional deep extraction from full papers:

in_depth:
  - small_dataset: Short explanation if fewer than ~1000 samples OR authors mention limited data.

(back to top)

📌 Usage

Run the CLI

research_agent analyze

You will be prompted to:

• Select a review type: meta or systematic
• Enter a search query (or load from config.yaml)
• Provide a screening objective (if filtering is enabled)

Example

research_agent analyze \
  --task meta \
  --objective "AI for diagnosing UTIs in dogs" \
  --filter-papers True \
  --max-results 100 \
  --output-dir ./output

Outputs

• paper_screenings.json: Results of the screening step

{
    "relevant": false,
    "reason": "The study does not involve the application of AI techniques such as machine learning or deep learning",
    "confidence": "High",
    "pmc": "PMC9554590",
    "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9554590/",
    "title": "Contrast-enhanced ultrasound features of focal pancreatic lesions in cats",
    "Author": "Silvia Burti",
    "year": "2022",
    "journal": "Frontiers in Veterinary Science"
  }

• paper_summaries.json: Meta-analysis or systematic review output

  {
    "meta_data": {
      "pmc": "PMC11271534",
      "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11271534/",
      "title": "Exploring deep learning strategies for intervertebral disc herniation detection on veterinary MRI",
      "journal": "Scientific Reports",
      "first_author": "Shoujin Huang",
      "year": "2024"
    },
    "body_data": {
      "small_dataset": "The study used 487 MRI images from 213 dogs, which is relatively small for deep learning applications.",
    }
  }

(back to top)

🗺 Roadmap

• Paper screening and summarization
• Configurable metadata fields
• Support local PDFs
• Interactive Q&A over papers
• Semantic Scholar and ArXiv support

(back to top)

🧾 License

Distributed under the MIT License. See LICENSE for details.

(back to top)