---
title: AI Literature Review System
emoji: 📚
colorFrom: blue
colorTo: green
sdk: gradio
sdk_version: 5.9.1
app_file: app.py
pinned: false
license: mit
---

# 📚 AI Literature Review System

An AI-powered literature review application that uses multiple specialized reviewers to analyze research papers. Built with Gradio and powered by OpenAI-compatible APIs.

## 🌟 Features

- **Multi-Agent Review System**: Three specialized AI reviewers with different perspectives:
  - **Experimentalist**: Focuses on methodology and experimental rigor
  - **Impactist**: Evaluates potential impact and field significance
  - **Novelty Seeker**: Assesses originality and innovation

- **Comprehensive Analysis**: Reviews papers across multiple dimensions:
  - Originality & Novelty
  - Technical Quality & Soundness
  - Clarity & Presentation
  - Significance & Impact
  - Contribution to the field

- **PDF Upload**: Simply upload your research paper and get instant feedback
- **MarkItDown Integration**: Advanced PDF text extraction for better accuracy
- **Semantic Scholar Integration**: Find and compare related papers
- **Detailed Feedback**: Receive strengths, weaknesses, questions, and actionable suggestions
- **Academic-Standard Scoring**: Based on top-tier conference review standards (NeurIPS-style)

## 🚀 Quick Start

### Local Installation

1. Clone the repository:
```bash
git clone https://huggingface.co/spaces/syaikhipin/PaperReview
cd PaperReview
```

2. Install dependencies:
```bash
pip install -r requirements.txt
```

3. Set up environment variables:
```bash
cp .env.example .env
# Edit .env with your API credentials
```

4. Run the application:
```bash
python app.py
```

### Hugging Face Spaces

Simply visit the space and configure your API settings in the UI or set secrets in Space settings:
- `OPENAI_API_KEY`: Your API key
- `OPENAI_BASE_URL`: API endpoint (optional)
- `MODEL_NAME`: Model identifier (optional)

## 📖 How to Use

1. **Configure API**: Enter your API credentials in the UI or use environment variables
2. **Upload Paper**: Upload your research paper in PDF format
3. **Enable Semantic Scholar** (optional): Search for related papers
4. **Review**: Click "Review Paper" and wait for sequential analysis (3-6 minutes)
5. **Analyze Results**: Review detailed feedback from all three reviewers

## 📊 Understanding the Scores

The system provides scores on a 1-10 scale:
- **9-10**: Award Quality / Strong Accept
- **7-8**: Accept
- **5-6**: Borderline
- **3-4**: Borderline Reject
- **1-2**: Reject

Each reviewer evaluates:
- **Soundness** (1-4): Technical quality
- **Presentation** (1-4): Writing quality
- **Contribution** (1-4): Overall contribution
- **Originality** (1-4): Novelty of ideas
- **Quality** (1-4): Research quality
- **Clarity** (1-4): Clarity of presentation
- **Significance** (1-4): Importance of results
- **Confidence** (1-5): Reviewer confidence level
- **Overall** (1-10): Overall assessment

## 🏗️ Project Structure

```
aireviewer/
├── app.py              # Main Gradio application
├── agents.py           # Multi-agent review system
├── requirements.txt    # Python dependencies
├── README.md          # This file
├── .env.example       # Environment variables template
└── .gitignore         # Git ignore rules
```

## 🛠️ Technical Details

### Multi-Agent Architecture

The system implements three specialized reviewer agents, each with a distinct persona:

1. **Experimentalist Reviewer**
   - Emphasizes experimental design and methodology
   - Looks for rigorous evaluation and clear insights
   - Questions reproducibility and statistical significance

2. **Impact-Focused Reviewer**
   - Evaluates potential field impact
   - Assesses practical applications
   - Considers broader implications

3. **Novelty-Focused Reviewer**
   - Seeks original contributions
   - Evaluates creative approaches
   - Identifies incremental vs. breakthrough work

### Review Process

1. **PDF Text Extraction**: Uses MarkItDown for high-quality text extraction
2. **Sequential Multi-Agent Review**: Each agent independently evaluates the paper one at a time
   - Rate limited to 1 request per second to avoid API concurrency issues
   - Sequential processing ensures consistent quality and respects API limits
3. **Scoring Aggregation**: Weighted scoring across multiple criteria
4. **Feedback Generation**: Structured feedback with JSON parsing
5. **Related Papers**: Semantic Scholar API integration with rate limiting (1 req/sec)

### API Compatibility

The system uses OpenAI-compatible APIs, supporting:
- OpenAI (GPT-4, GPT-3.5, etc.)
- Azure OpenAI
- Custom endpoints (LocalAI, vLLM, LiteLLM, etc.)
- Any OpenAI-compatible inference server

## 🔒 Privacy & Security

- API keys can be provided via UI or environment variables
- Uploaded PDFs are processed in memory and not permanently stored
- All processing happens on your infrastructure
- Semantic Scholar searches are anonymous

## 🌐 Deployment

### Hugging Face Spaces

This app is deployed on Hugging Face Spaces. To deploy your own:

1. Create a new Space on Hugging Face
2. Select "Gradio" as SDK
3. Upload all files from this repository
4. Configure secrets in Space settings:
   - `OPENAI_API_KEY` (required)
   - `OPENAI_BASE_URL` (optional, defaults to OpenAI)
   - `MODEL_NAME` (optional, defaults to gpt-3.5-turbo)
   - `SEMANTIC_SCHOLAR_API_KEY` (optional but recommended for better rate limits)

### Docker

```dockerfile
FROM python:3.10-slim
WORKDIR /app
COPY requirements.txt .
RUN pip install -r requirements.txt
COPY . .
CMD ["python", "app.py"]
```

## 📝 Environment Variables

```bash
# Required
OPENAI_API_KEY=your-api-key-here

# Optional
OPENAI_BASE_URL=https://api.openai.com/v1
MODEL_NAME=gpt-4
SEMANTIC_SCHOLAR_API_KEY=your-semantic-scholar-key
```

**Note**: All API keys should be stored as **secrets** in Hugging Face Spaces settings or in a local `.env` file (never commit .env to git).

## ⚠️ Notes

- Reviews are generated **sequentially** (one at a time) with 1-second delays between API calls
- **Rate Limiting**: Both LLM and Semantic Scholar APIs are rate-limited to 1 request/second to avoid concurrency issues
- Processing time: 3-6 minutes depending on paper length and API response time
- Ensure your PDF contains extractable text (not scanned images)
- Semantic Scholar API: Using an API key provides higher rate limits (recommended)
- The system provides automated feedback; human review is still recommended

## 🤝 Contributing

Contributions are welcome! Feel free to:
- Report bugs
- Suggest new features
- Submit pull requests
- Improve documentation

## 📄 License

This project is open source and available under the MIT License.

## 🙏 Acknowledgments

- Based on multi-agent research frameworks
- Inspired by academic peer review processes
- Uses MarkItDown for PDF processing
- Integrates with Semantic Scholar API
- Built with Gradio for easy deployment

## 📧 Contact

For questions or feedback, please open an issue on the repository.

---

**Live Demo**: https://huggingface.co/spaces/syaikhipin/PaperReview