🌌 ASTREA

A Space Biology Research Engine

Winner (1st Place) — NASA International Space Apps Challenge 2025 (Ghaziabad Local Hackathon) 🏆

ASTREA is an AI-powered research engine designed to help scientists and researchers explore large-scale space biology literature efficiently. It combines semantic search, knowledge graph reasoning, and LLM-powered synthesis to uncover hidden relationships across NASA bioscience studies.

Built by Team Andromeda during the NASA Space Apps Challenge 2025.

---

🚀 Problem Statement

NASA’s space biology research produces vast amounts of unstructured and semi-structured data—research papers, experimental reports, tables, and images. However:

• Relevant insights are scattered across thousands of documents
• Cross-paper relationships are difficult to identify
• Traditional keyword search fails to capture semantic meaning

As a result, valuable connections remain hidden, slowing scientific discovery.

---

💡 Solution Overview

ASTREA introduces a hybrid retrieval and reasoning system that:

• Scrapes and ingests bioscience documents

• Extracts structured knowledge (entities & relationships)

• Stores information in both:

  • a Vector Database (semantic similarity)
  • a Knowledge Graph (explicit relationships)

• Answers user queries by reasoning over both representations

This allows ASTREA to go beyond simple document retrieval and generate context-aware, insight-driven responses.

---

🧠 System Architecture

High-level flow:

1. Document Ingestion

   • Web scraping (HTML pages)
   • PDF document loading

2. Chunking & Preprocessing

   • Intelligent text chunking with overlap
   • Metadata enrichment (source, timestamp, type)

3. Dual Storage Pipeline

   • Vector DB (Pinecone) Stores embeddings for semantic retrieval
   • Knowledge Graph (Neo4j) Stores extracted entities and relationships as graph triples

4. Query Processing

   • User query is embedded and matched against vector DB
   • Related entities and edges fetched from Neo4j

5. LLM Reasoning Layer

   • Retrieved chunks + graph context are combined
   • LLM generates a grounded, reasoned response

This hybrid approach enables both semantic similarity and relational reasoning.

---

✨ Key Features

• 🔍 Semantic Search across 600+ NASA bioscience studies
• 🧠 Hybrid RAG Architecture (Vector + Knowledge Graph)
• 🕸️ Entity & Relationship Extraction using LLMs
• 📊 Neo4j Knowledge Graph for relational exploration
• 📦 Pinecone Vector Store for scalable similarity search
• 🧩 Intelligent Chunking for high-quality retrieval
• 📄 PDF + Web Content Support
• 🚀 Multiple Interfaces: REST API, CLI, and programmatic usage

---

🛠️ Tech Stack

• Backend: Node.js, Express.js
• Web Scraping: Axios, Cheerio
• LLMs & Embeddings: OpenAI / Gemini (configurable)
• Vector Database: Pinecone
• Knowledge Graph: Neo4j
• Document Processing: Custom chunking pipeline
• APIs: RESTful endpoints

---

📦 Installation & Setup

Prerequisites

Create a .env file in the root directory:

GEMINI_API_KEY=your_api_key
OPENAI_API_KEY=your_api_key
PINECONE_INDEX_NAME=your_index_name
NEO4J_URI=your_neo4j_uri
NEO4J_USERNAME=your_username
NEO4J_PASSWORD=your_password
PORT=8080

Install dependencies

npm install

---

▶️ Usage

Start the server

npm start
# or
npm run dev

---

REST API Endpoints

Scrape a single URL

POST /scrape/url

{
  "url": "https://example.com/research-paper"
}

Scrape multiple URLs

POST /scrape/urls

{
  "urls": ["https://site1.com", "https://site2.com"]
}

Ask a question

POST /chat

{
  "question": "How does microgravity affect gene expression in mice?"
}

---

CLI Tool

npm run scraper-cli

The CLI supports:

• Single / multiple URL ingestion
• Progress tracking
• Error diagnostics

---

🧪 How ASTREA Thinks (Why It’s Different)

Traditional RAG

Query → Retrieve Documents → LLM → Answer

ASTREA’s Hybrid Reasoning

Query
  ↓
Vector Retrieval (Pinecone)
  +
Graph Traversal (Neo4j)
  ↓
Reasoning Layer
  ↓
LLM
  ↓
Insightful Answer

This allows ASTREA to:

• Infer relationships
• Connect experiments across papers
• Provide more grounded, explainable answers

---

🏆 Achievements

• 🥇 Winner (1st Place) — NASA International Space Apps Challenge 2025 Ghaziabad Local Hackathon
• Built by Team Andromeda

Team Members:

• Anchita Jain
• Ankit Gupta
• Ashutosh Kumar Singh
• Ananya Patel
• Akhand Pratap Singh
• Lakshya Pratap Singh

---

🎥 Demo & Presentation

• 📽️ Demo Video: https://drive.google.com/file/d/1sU4GM5Dd47xY-x1pKavkodyNyzcAPI7h/view
• 📄 Pitch Deck & Architecture: Included in repository

---

📜 License

This project is licensed under the ISC License.

---

🙌 Acknowledgements

Special thanks to:

• NASA Space Apps Challenge
• Innogeeks & KIET Group of Institutions
• Judges and mentors for valuable feedback and guidance