An AI-powered chatbot for SETI Institute citizen scientists, delivering instant, level-specific answers to astronomy and Unistellar telescope queries through a RAG-based system.
Techtinium developed an ASTRO LLM, an AI-powered knowledge companion for the SETI Institute and Unistellar’s citizen scientists, automating responses to astronomy-related queries while supporting varying levels of user expertise.
About SETI Institute & Senior Citizen Scientist Program
The SETI Institute engages in the scientific search for life beyond Earth, involving over 100 scientists and specialists across research, education, and outreach. Through collaborations with Unistellar, SETI mobilizes a global network of more than 8,000 citizen scientists using the Enhanced Vision Telescope (eVscope) to contribute to real scientific campaigns.
Citizen scientists frequently asked questions about observing campaign-specific targets, including asteroids, comets, variable stars, and transient astronomical events. Previously, these questions were routed via Slack to SETI scientists, creating a significant workload and slowing response times.
The Challenge
- Automate responses to questions from citizen scientists without reducing engagement.
- Provide timely and accurate information for both general astronomy and Unistellar target and observation specific queries.
- Support multiple expertise levels, from beginner to advanced, including providing equations and detailed observational instructions when appropriate.
- Integrate with existing systems and observations, initially limited to data through 2024.
Solution Engineered By Techtinium
- RAG-Based Chatbot Development: Built ASTRO LLM using Python, Flask, and Google Vertex AI’s RAG capabilities to source information from structured documents.
- Automatic knowledge extraction: Capability to look up open source websites like the ADS, arXiv, and wikipedia for queries about novel targets and integrating the acquired knowledge into the RAG system.
- Dynamic Level-Based Responses: Users can select beginner or advanced modes to tailor answer complexity. Advanced answers include equations and technical data, while beginner answers provide simplified explanations.
- Campaign-Specific Knowledge Integration: Includes observation datasets for Unistellar telescope campaigns (asteroids, exoplanets, comets, planetary defense, and transient events).
Future Plan of Action
- Fully Agentic Reasoning: ASTRO LLM will autonomously determine the scope, campaign, and target involved in the user’s query along with the level of expertise, and required data extraction sources without explicit input.
- Web Search Integration: The system will possess the capability to pull information from open source websites to provide up-to-date answers on celestial events and observations.
- Expanded Dataset Updates: Beyond 2024, the model will continuously integrate new research papers, targets discovered, and observations made by citizen scientists.
- Enhanced User Interaction: Improved UI and conversational features to make it seamless for citizen scientists to explore astronomy topics.
Impact
- Reduces scientist workload by automating routine query responses
- Empowers citizen scientists with immediate, accurate answers using knowledge from open source and peer-reviewed data sources
- Supports scalable participation in astronomy campaigns
- Lays the foundation for real-time, intelligent guidance in observational astronomy.
