AI-powered research paper analysis system

Inspiration

The process of analyzing dense academic research is slow and manual. We were inspired by AWS Bedrock to build a tool that automates this process, turning static papers into interactive knowledge hubs and acting as a true AI research assistant for students, scientists, and professionals.

What it does

ResearchAI is a full-stack platform that uses AWS Bedrock to supercharge research paper analysis. After uploading a PDF, users can:

• Chat with the Document: Ask natural language questions and get accurate, source-cited answers using a Retrieval-Augmented Generation (RAG) pipeline.
• Generate Comprehensive Summaries: Instantly create a structured 9-section analysis covering the paper's methodology, results, and implications.
• Create Interactive Quizzes: Test and reinforce understanding with AI-generated multiple-choice questions and detailed explanations.
• Visualize Concepts: Generate an interactive mind map to visually explore the paper's structure and key concepts.
• Conduct Autonomous Research: Deploy an AI agent to perform web research on any topic, which then generates a comprehensive, downloadable report.

How we built it

We designed a modern, scalable architecture to deliver a seamless user experience.

• AI & Cloud:
  • AWS Bedrock: Used Amazon Nova Pro v1 for reasoning and generation, and Amazon Titan Embeddings v2 for high-quality semantic search.
  • AWS S3: For durable, scalable storage of all documents.
  • Vector Database: Pinecone for fast and efficient similarity search.
• Backend: FastAPI for high-performance, asynchronous processing of documents and embeddings.
• AI Orchestration: LangChain and LangGraph to build our RAG pipeline and the sophisticated, multi-step autonomous research agent.
• Frontend: React 18 and Tailwind CSS for a responsive and modern user interface.

Challenges we ran into

1. Agentic Workflow Complexity: Building a reliable autonomous agent required more than a simple LLM call. Solution: We implemented LangGraph to create a stateful, cyclical workflow, allowing the agent to plan, search, and reason over multiple steps.

2. Guaranteeing Factual Accuracy: To prevent AI hallucinations, we needed to ground all responses in the source text. Solution: We engineered a robust RAG pipeline using high-quality Amazon Titan Embeddings, ensuring the context provided to Amazon Nova Pro was precise and relevant.

Accomplishments that we're proud of

• Deep AWS Bedrock Integration: We successfully utilized multiple Bedrock models for their specific strengths, demonstrating a sophisticated, enterprise-grade use of the AWS AI ecosystem.
• Fully Autonomous Research Agent: Building a functional agent that can independently conduct research and compile a detailed report was a major achievement.
• Multi-Modal Analysis: We are proud of creating a platform that offers diverse analysis tools (chat, summary, quiz, mind map) from a single document upload, providing holistic value to the user.

What we learned

• The power of managed services like AWS Bedrock to rapidly prototype and deploy complex AI systems without managing infrastructure.
• The critical importance of a well-engineered RAG pipeline, where high-quality embeddings from Amazon Titan are essential for factual accuracy.
• The potential of agentic workflows using LangGraph to solve complex, multi-step problems that go beyond simple Q&A.

What's next for AI-powered research paper analysis system

Our vision is to evolve ResearchAI into a collaborative, multi-user platform. Key next steps include:

• Implementing user authentication and a shared paper library.
• Developing a multi-paper comparison feature.
• Integrating with enterprise knowledge bases using Amazon Kendra.
• Building a CI/CD pipeline for automated testing and deployment.
• Leveraging AWS Bedrock AgentCore for workflow orchestration

Built With

• amazon-nova-pro
• amazon-titan-embeddings
• amazon-web-services
• aws-bedrock
• boto3
• fastapi
• framer-motion
• langchain
• langgraph
• mistral-ai-api
• node.js
• perplexity-api
• pinecone
• python
• react
• tailwind-css
• tavily-api
• vite