Skip to content

README.md

Latest commit

 

History

History
229 lines (155 loc) · 4.82 KB

README.md

File metadata and controls

229 lines (155 loc) · 4.82 KB

📚 Deep Read Agent (PDF)

An Open-Source AI Research Assistant
Automatically discover papers → extract full PDFs → generate deep-reading insights → sync to Notion.

🚀 Overview

Deep Read Agent is an open-source research automation tool that transforms academic paper discovery into a structured, AI-powered workflow.

It automatically:

  • 🔍 Searches recent papers (arXiv + optional HuggingFace)
  • 📊 Ranks them using recency, citations, and keyword relevance
  • 📄 Downloads and extracts full PDF content
  • 🧠 Generates structured deep-reading analysis using an LLM
  • 📝 Saves structured insights into a Notion database (with deduplication)

This project turns scattered research reading into a scalable knowledge system.

🧠 Example Output Structure

Each paper produces:

  • 🔥 Recommendation Level
  • 📌 Deep Summary
  • 🧪 Methods & Experimental Design
  • 💡 Key Innovations & Contributions
  • 🛠 Improvement Ideas (If I Were to Extend This Work)

🏗 Architecture

flowchart TD
    A[User Topic Input] --> B[ask_config]
    B --> C[discover_and_extract]
    C --> D[Search arXiv]
    D --> E[Rank Papers]
    E --> F[Download PDF]
    F --> G[Extract Full Text]
    G --> H[Chunk Processing]
    H --> I[LLM Deep Analysis]
    I --> J[save_to_notion]
    J --> K[Notion Database]
Loading

📦 Core Modules

1️⃣ ask_config()

Configures:

  • Time range (7 / 30 / 90 / 180 days)
  • Number of papers to deeply analyze
  • Topic keywords

Default configuration:

{
  "time_range_days": 30,
  "deep_read_count": 2
}

2️⃣ discover_and_extract()

Responsibilities:

  • Search arXiv
  • (Optional) Fetch citation counts
  • Rank papers
  • Download PDFs
  • Extract full text
  • Chunk text (≤10 chunks)

Returns structured paper objects ready for analysis.

3️⃣ save_to_notion()

Saves structured results to a Notion database.

Required database fields:

Field Name Type

Title title Paper URL url Recommendation select Deep Summary rich_text Methods & Experiments rich_text Key Innovations rich_text Improvement Ideas rich_text

🧮 Ranking Algorithm

Score combines:

  • Keyword match weight
  • Recency boost
  • log10(citationCount)

Formula:

score = keyword_score + recency_score + log10(citationCount)

🛠 Installation

Node.js >= 18 recommended

npm install

Dependencies:

  • node-fetch
  • pdf-parse
  • zod

▶️ Usage

Step 1: Configure

await ask_config()

Step 2: Discover & Extract

await discover_and_extract({
  topic: "LLM inference efficiency",
  time_range_days: 30,
  deep_read_count: 2
})

Step 3: Run LLM Analysis

Generate structured deep insights using your preferred LLM.

Step 4: Save to Notion

await save_to_notion({
  notion_api_key: "...",
  notion_database_id: "...",
  items: [...]
})

🔐 Security

  • No API keys are stored
  • No credential persistence
  • Runtime-only authentication
  • No sensitive data logging

🌟 Why This Project Matters

Traditional research workflows:

  • Manual filtering
  • Surface-level reading
  • Disorganized notes

Deep Read Agent enables:

  • Automated discovery
  • Full-text understanding
  • Structured knowledge capture
  • Long-term research intelligence

This is a foundation for building a scalable AI Research OS.

📈 Roadmap

  • OpenAlex integration
  • Local PDF ingestion
  • Embedding-based deduplication
  • Automated research roadmap generation
  • Multi-database export support

📜 License

MIT

⭐ Support

If this project helps your research:

Star ⭐
Fork 🍴
Contribute 🚀

Let's build open research intelligence.