ASTREA — Space Biology Research Engine

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

ASTREA is an AI-powered research engine that helps scientists and researchers explore NASA space biology literature using semantic search, knowledge graphs, and Large Language Models (LLMs). It enables users to discover meaningful relationships across hundreds of research papers and receive context-aware, evidence-based answers.

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

Problem Statement

NASA's space biology research consists of hundreds of research papers, reports, and experimental studies. Finding relevant information across these documents is difficult because:

Research is spread across numerous documents.
Traditional keyword search cannot understand semantic meaning.
Relationships between studies are difficult to identify.
Researchers spend significant time manually searching for relevant information.

Solution

ASTREA combines Retrieval-Augmented Generation (RAG), vector search, and knowledge graph reasoning to provide intelligent answers from scientific literature.

The system:

Collects and processes NASA bioscience documents.
Extracts important entities and relationships.
Stores semantic embeddings in a vector database.
Builds a knowledge graph for relational reasoning.
Combines both retrieval methods with an LLM to generate accurate, context-aware responses.

Key Features

Semantic search across 600+ NASA bioscience studies.
Hybrid RAG architecture using vector search and knowledge graphs.
Intelligent document chunking for improved retrieval.
Entity and relationship extraction.
Knowledge graph exploration using Neo4j.
Context-aware AI-generated responses.
Support for both PDF documents and web-based research articles.

System Architecture

The overall workflow consists of:

Document ingestion from web pages and PDF files.
Intelligent document chunking and preprocessing.
Embedding generation for semantic search.
Storage in Pinecone for vector similarity search.
Entity extraction and relationship storage in Neo4j.
Hybrid retrieval using both vector search and graph traversal.
Response generation using an LLM with retrieved context.

This architecture combines semantic similarity with explicit relationship reasoning to improve answer quality.

Tech Stack

Backend

Node.js
Express.js

AI & LLM

LangChain
OpenAI API
Google Gemini

Databases

Pinecone
Neo4j

Document Processing

Axios
Cheerio
Custom document chunking pipeline

My Contributions

Designed and implemented the Retrieval-Augmented Generation (RAG) pipeline.
Built the document ingestion and preprocessing pipeline.
Integrated Pinecone for semantic retrieval.
Designed the Neo4j knowledge graph for entity relationships.
Implemented hybrid retrieval combining vector search and graph reasoning.
Integrated LLMs to generate context-aware research responses.

Achievements

Winner (1st Place) — NASA International Space Apps Challenge 2025 (Ghaziabad Local Hackathon).
Indexed and processed over 600 NASA bioscience research papers.
Built a hybrid AI search engine combining vector retrieval with knowledge graph reasoning.

Acknowledgements

Special thanks to NASA Space Apps Challenge, Innogeeks, KIET Group of Institutions, mentors, and judges for their guidance and support throughout the competition.