🧩 Biotessera: A Space Biology Knowledge Engine

Biotessera is an AI-powered agent system developed for the NASA Space Apps Challenge 2025. It transforms hundreds of NASA publications on space biology into a navigable and interactive knowledge mosaic, enabling scientists and mission planners to find precise, synthesized answers backed by source data.

🎯 The Problem

NASA's decades of space biology research represent a vast and invaluable resource. However, this knowledge is spread across hundreds of documents, making it difficult to quickly find specific, cross-referenced information needed for planning future long-duration missions.

💡 The Solution

Biotessera acts as an intelligent research assistant. It uses a multi-agent architecture (Coordinator/Worker pattern) to address this challenge:

1. Data Preparation: An offline pipeline processes 607 full-text publications, extracts metadata, and generates vector embeddings for every text fragment (a "tessera"). These are stored in a local vector database (TesseraStore).
2. Retrieval: The main agent, TesseraConductor, receives a user's question and delegates tasks to specialized tools:
   • TesseraMiner: Searches the local TesseraStore for the most relevant and diverse text fragments from the 607 publications, utilizing Maximal Marginal Relevance (MMR) for enhanced result quality.
   • DataFinder: Performs real-time searches on the NASA Open Science Data Repository (OSDR) to find related raw datasets.
3. Synthesis: The TesseraConductor gathers all retrieved information and uses a Large Language Model (LLM) to generate a single, coherent, and sourced answer.

✨ Features

• Natural Language Q&A: Ask complex questions in plain English.
• Multi-Source Answers: The agent can combine information from its internal knowledge base (TesseraStore) and external NASA databases (OSDR).
• Source-Backed: Every statement in the generated answer is based on the retrieved data, preventing AI "hallucinations".
• Modular Architecture: Easily extendable with new tools to search other databases.

🛠️ Tech Stack

• Backend: Python
• AI/ML: LangChain, Google Gemini (LLM & Embeddings)
• Vector Database: ChromaDB
• Data Processing: Pandas, BeautifulSoup
• UI: Streamlit

☁️ Data Hosting

The vector database for this project (tesserastore_db) is approximately 808 MB, which exceeds GitHub's file size limits. To ensure the live Streamlit application can be deployed, the database is compressed and hosted on Hugging Face Datasets.

• Dataset Link: vero-code/biotessera-database

The application automatically downloads and unpacks this database on its first run in a new environment.

🚀 How to Run

1. Clone the repository:

   git clone https://github.com/vero-code/biotessera.git
   cd biotessera

2. Set up the environment:

   python -m venv venv
   source venv/bin/activate  
   # On Windows: .\venv\Scripts\activate
   pip install -r requirements.txt

3. Add your API Key:
   • Create a .env file in the root directory.
   • Add your Google AI API key to it: GOOGLE_API_KEY="YOUR_API_KEY_HERE"
4. Run the application:

   streamlit run app.py

🛰️ NASA Data & Resources Used

This project utilizes the following official NASA resources:

• A list of 608 full-text open-access Space Biology publications: The primary knowledge base for the Biotessera agent.
• NASA Open Science Data Repository (OSDR): Used by the DataFinder tool to perform real-time searches for raw experimental datasets.

🙏 Acknowledgments

Special thanks to the organizers, mentors, and the entire community of the NASA International Space Apps Challenge 2025 for making this event possible.

📜 License

This project is licensed under the MIT License.