AI Research Paper Search Agent

An intelligent agent that searches, analyzes, and ranks arXiv research papers using LangGraph, LangChain, OpenAI GPT-4, and DuckDuckGo.

The AI agent helps researchers find relevant papers by understanding search intent and providing iterative refinement of search results.

The Story Behind the Project

The Problem

As AI researcher and developer, I frequently found myself drowning in a sea of academic papers. The traditional process was time-consuming and frustrating:

• Spending hours scanning through papers to find relevant research
• Reading lengthy abstracts that might not align with my needs
• Manually tracking and organizing papers across different topics
• Missing important papers due to poor keyword matches
• Getting stuck in research rabbit holes that led nowhere

The Solution

This frustration led to the development of the AI Research Paper Search Agent that can work as AI Research Assistant.

The goal was simple: create a tool that thinks like a researcher but works at the speed of a computer.

The AI agent doesn't just find papers – it understands them, explains their relevance, and helps refine the search based on what you actually need.

Key innovations:

• Intent Understanding: Unlike traditional keyword searches, the AI agent understands research goals and matches papers based on actual relevance
• Intelligent Refinement: The system learns from your feedback and refines searches iteratively
• Comprehensive Analysis: Each paper is automatically summarized and evaluated for relevance to your specific research goals
• Time Efficiency: What used to take days can now be accomplished in minutes

Development Journey

The project evolved through several stages:

1. Initial Prototype: Started with a simple script using OpenAI's API to summarize papers
2. Architecture Evolution:
   • Added DuckDuckGo integration for broader paper discovery
   • Used LangGraph to implement structured AI workflows
   • Developed a ranking system based on relevance scoring
3. User Experience:
   • Created an intuitive web interface with Gradio
   • Added interactive refinement capabilities
   • Implemented session management for iterative searches
4. Optimization:
   • Improved rate limiting and error handling
   • Added asynchronous processing for better performance
   • Enhanced abstract extraction reliability

Use Cases

The AI agent can be used for:

• Academic Research: PhD students using it to conduct comprehensive literature reviews
• Industry R&D: Companies tracking technological developments in their field
• Cross-Domain Research: Researchers exploring intersections between different fields
• Education: Students getting quick overviews of complex research areas

Features

• Intelligent Paper Search: Searches arXiv papers using DuckDuckGo with date range filtering
• Automated Analysis: Generates paper summaries and explains relevance to search intent
• Smart Ranking: Ranks papers based on relevance to research goals using LLM
• Interactive Refinement: Allows users to refine search intent based on initial results
• Web Interface: Clean Gradio interface for easy interaction
• Iterative Search: Supports up to 3 iterations of intent refinement per search session

Technical Stack

• LangGraph: For building the AI agent workflow
• OpenAI GPT-4: For paper analysis and ranking
• DuckDuckGo Search: For paper discovery
• Gradio: For web interface
• Beautiful Soup: For paper abstract extraction
• Pydantic: For data validation and modeling

Installation

1. Clone the repository:

git clone https://github.com/alexey-tyurin/ai-agent.git
cd ai-agent

2. Install required packages:

pip install -r requirements.txt

3. Set up environment variables:

export OPENAI_API_KEY="your-api-key"

Optional for debugging:

export LANGCHAIN_API_KEY="your-langchain-key"
export LANGCHAIN_TRACING_V2="true"
export LANGCHAIN_PROJECT="ai-agent"

Usage

1. Start the application:

python main.py

2. Access the web interface at http://localhost:7860

3. Enter search parameters:

   • Keywords: Search terms for papers
   • Intent: Research goal or specific interest
   • Date range: Filter papers by publication date
   • Number of results: How many papers to analyze (1-20)

4. Review results and optionally:

   • Continue search with refined intent
   • Stop search when satisfied with results

Architecture

Components

1. Search Pipeline:

   • Paper search using DuckDuckGo
   • Abstract extraction from arXiv
   • Paper summarization
   • Relevance analysis
   • Result ranking

2. State Management:

   • Session tracking
   • State persistence
   • Iterative refinement

3. Models:

   • Paper: Basic paper metadata
   • PaperSummary: Includes AI-generated summary
   • PaperExplanation: Adds relevance explanation
   • RankedPaper: Includes relevance score
   • SearchQuery: Search parameters and session info
   • AgentState: Workflow state tracking

Workflow

1. Search: Find papers matching keywords and date range
2. Summarize: Generate concise summaries
3. Explain: Analyze relevance to search intent
4. Rank: Score papers based on relevance
5. Review: Present results for user feedback
6. Refine: Optionally continue with new intent

Error Handling

• Rate limiting protection for DuckDuckGo searches
• Validation for date formats and ranges
• Error handling for paper abstract extraction
• Score normalization for paper rankings

Development Notes

• Maximum 3 iterations per search session
• Scores normalized to 0-10 range
• Configurable result limit (1-20 papers)
• Asynchronous processing for LLM operations

Future Improvements

• Enhance explaining relevance to search intent algorithm
• Enhance ranking algorithm
• Add result caching
• Add saving results to files functionality
• Add export functionality for results
• Add support for additional paper repositories
• Add paper clustering by topic

Screenshots of the flow

1) Initial screen

2) First results

3) Changing intent

4) Continue search with new intent

5) Final results with updated intent

Acknowledgments

References

https://www.deeplearning.ai/the-batch/how-agents-can-improve-llm-performance/

https://www.deeplearning.ai/the-batch/agentic-design-patterns-part-2-reflection

https://www.deeplearning.ai/the-batch/agentic-design-patterns-part-3-tool-use/

https://www.deeplearning.ai/the-batch/agentic-design-patterns-part-4-planning/

https://www.deeplearning.ai/the-batch/agentic-design-patterns-part-5-multi-agent-collaboration/

https://academy.langchain.com/courses/intro-to-langgraph

https://langchain-ai.github.io/langgraph/

https://smith.langchain.com/

My certificate for LangGraph

https://academy.langchain.com/certificates/oylo2acrsw

Contact Information

For any questions or feedback, please contact Alexey Tyurin at altyurin3@gmail.com.

License

This project is licensed under the MIT License - see the LICENSE file for details.