AstroBio Explorer

AstroBio Explorer is our AI-powered research assistant built for the NASA Space Apps Challenge 2025(Build a Space Biology Knowledge Engine). It helps anyone explore NASA’s space biology research in a simple, conversational and interactive way.

---

🌟 Inspiration

NASA has been studying how life behaves beyond Earth for decades. From how plants grow in space to how microgravity affects the human body, the knowledge is incredible. But it’s buried inside hundreds of research papers that are hard to search and connect.

We wanted to make that exploration feel natural, like having a conversation with a scientist who knows exactly where to look. That’s why we built AstroBio Explorer.

---

🧭 What It Does

AstroBio Explorer turns complex scientific literature into a friendly research companion. Here’s what you can do inside it:

• 🧠 Ask research questions in plain language and get clear, well-grounded answers from NASA bioscience papers
• 🔗 See inline citations with clickable links to original publications
• 📄 Explore related Google Scholar papers using SerpAPI
• 🖼 View relevant images tied to your question in a sidebar
• 💬 Keep multiple research threads open using multi-session chat

It doesn’t just give you a block of text. It brings together papers, citations, images and related work so you can truly explore a topic.

---

🎥 Demo

👉 Watch the Demo on YouTube

---

🛠 How We Built It

We structured AstroBio Explorer into three main parts working together.

🧠 Backend (RAG + AI)

• Built with FastAPI, LangChain, and Gemini 2.5 Flash
• Parsed and chunked 608 NASA bioscience articles at the sentence level
• Generated dense and sparse embeddings and stored them in Qdrant for fast retrieval
• Used a small agent system to combine the RAG pipeline with Google Scholar and image search, merging multiple sources for each response

🖥 Frontend (UI/UX)

• Built with Next.js and Tailwind CSS
• Clean chat interface with SCSS and Haml animations for a smooth space-themed feel
• Citations appear inline inside chat bubbles
• Images are shown in a sidebar for visual context using Image Fetch API(Serp API)
• Includes a Google Scholar button to fetch related papers(Serp API)
• Supports multiple chat sessions so users can explore different topics simultaneously

🧱 Data Layer

• Vector DB: Qdrant, storing each chunk with metadata like paper ID, section, and URL for precise citation linking
• Context + session storage: For multi-turn conversations and persistent threads

---

🪄 Pipeline Overview

---

🚧 Challenges We Ran Into

• Finding the right chunk size to keep answers precise without slowing the system
• Placing inline citations at exactly the right spots in the generated text
• Coordinating Gemini, Qdrant and SerpAPI in a single pipeline
• Designing a UI that could handle citations, media and multiple threads without feeling cluttered

---

🏆 Accomplishments We're Proud Of

• Built a working research assistant that connects NASA data, Google Scholar and images in one conversational interface
• Added inline citations that link directly to real papers
• Enabled multi-session research chats for parallel exploration
• Designed a modular backend that can easily be extended with more tools later

---

📚 What We Learned

• How to combine different AI components into a smooth, useful experience
• The importance of chunking and retrieval when working with real scientific literature
• How to shape LLM outputs into structured answers that are actually useful for research
• How to design frontend components that gracefully handle citations, references and visual content

---

🚀 What’s Next for AstroBio Explorer

• A Research Mode with longer, more detailed answers and expanded context
• Visual knowledge graphs that show how concepts connect across studies
• Hypothesis generation and exportable research briefs
• Direct integration with NASA’s open data repositories for richer exploration

---

🧰 Built With

• Next.js
• Tailwind CSS
• SCSS + Haml animations
• FastAPI
• LangChain
• Gemini 2.5 Flash
• Qdrant
• SerpAPI (Google Scholar + Image Fetch)

---

🧪 Evaluation Overview

We tested AstroBio Explorer on real NASA space biology literature to measure how well it retrieves, cites and answers scientific questions.

• 181 questions
• 42 research papers
• Evaluation measured retrieval, answer quality and citation accuracy

Results

• Overall Score: 72% (Weighted across retrieval, answer quality, and citation accuracy)

Component	Score	Notes
Answer Quality	83%	Excellent factual alignment with expert answers
Citation Accuracy	70%	Good relevance and proper linking
Retrieval Quality	59%	Good but room for broader document recall

Formula:

Overall = 35% × Retrieval + 45% × Answer Quality + 20% × Citations

Performance by Question Type

Question Type	Score
Specific questions	79%
Comparative analysis	77%
Factual queries	73%
Complex reasoning	69%
Broad overviews	60%

AstroBio Explorer excels at targeted scientific questions, which is where researchers benefit most.

Top Performance Areas

• Medical case studies – 89%
• Space agriculture research – 86%
• Plant imaging systems – 88%
• FAIR data systems – 88%

---

🔐 Environment Setup

Create a .env file for the backend and a .env.local for the frontend. Do not commit these files to Git. If you ever push them by mistake, rotate the keys immediately.

Backend .env (create in backend/.env)

# LLM
GEMINI_API_KEY=your_gemini_api_key
MODEL_NAME=gemini-2.5-flash

# Vector DB
QDRANT_URL=https://your-qdrant-host:6333
QDRANT_API_KEY=your_qdrant_api_key
QDRANT_COLLECTION=space_bio_chunks

# External search
SERPAPI_API_KEY=your_serpapi_key

# CORS
ALLOW_ORIGINS=http://localhost:3000

Frontend .env.local (create in frontend/.env.local)

# Your API base for local dev
NEXT_PUBLIC_API_BASE_URL=http://localhost:8000

Git ignore tip

Make sure your root .gitignore includes:

.env
.env.local
**/.env
**/.env.local

---

🧭 How to Run the Project

1. Backend: create virtual environment

cd backend
python -m venv venv

2. Backend: activate virtual environment

macOS or Linux:

source venv/bin/activate

Windows PowerShell:

venv\Scripts\Activate.ps1

3. Backend: install dependencies

pip install -r requirements.txt

4. Backend: set environment variables from .env

Make sure backend/.env exists with your keys. If you use a process manager it will load automatically. For local shells, you can also export manually if needed.

macOS or Linux:

export $(grep -v '^#' .env | xargs)

Windows PowerShell:

Get-Content .env | ForEach-Object {
  if ($_ -and $_ -notmatch '^\s*#') {
    $name, $value = $_.Split('=',2)
    [Environment]::SetEnvironmentVariable($name.Trim(), $value.Trim())
  }
}

5. Backend: run the API

python main.py

The backend will be available at:

http://localhost:8000

6. Frontend: install dependencies

cd ../frontend
npm install

7. Frontend: run the dev server

npm run dev

The frontend will be available at:

http://localhost:3000

---

📂 Project Structure

.
├── backend        # FastAPI + LangChain + Gemini chatbot + RAG pipeline
├── frontend       # Next.js + Tailwind CSS UI
├── evaluation     # Evaluation scripts, metrics and results

---

👨‍🚀 Meet AstroBio Explorer

AstroBio Explorer is more than a chatbot. It’s a research companion designed to make space bioscience accessible and explorable. Whether you’re a scientist, a student or just curious about how life thrives beyond Earth, AstroBio Explorer lets you ask real questions and get meaningful, research-backed answers.