Yu-Yang-Li/StarWhisper

What it solves

StarWhisper provides a suite of AI-driven tools and frameworks to automate astronomical research and telescope operations. It addresses the challenges of managing complex observing schedules, classifying celestial objects (like variable stars and pulsars) from light curves, and estimating the age of supernova explosions to prioritize targets for observation.

How it works

The project is delivered as a "skill pack" containing 17 research skills that can be integrated into AI coding assistants like Codex or Cursor. It employs several specialized AI approaches:

• Observation Agents: An AI framework (StarWhisper Telescope) that manages end-to-end telescope operations, handling real-time disruptions like weather or hardware failures to decide whether to continue, insert new targets, or reschedule.
• Classification Models: Uses Conv1D-BiLSTM, Swin Transformers, and multimodal LLMs to classify light curves from Kepler/K2 data and identify pulsar candidates.
• Domain-Specific LLMs: StarWhisper 3 is a language model trained specifically for astronomical Q&A and code generation.
• Predictive Tools: The "SN Clock" system estimates the explosion age of supernova candidates to help astronomers identify the youngest candidates for urgent follow-up.

Who it’s for

Astronomers, astrophysicists, and AI researchers focusing on embodied intelligence for robotics (telescope control) and time-series analysis of astronomical data.

Highlights

• End-to-End Automation: Automates the loop from target selection to telescope execution via the NGSS (Near Galaxy Survey System).
• Multimodal Analysis: Applies LLMs and audio models to light curve classification with approximately 90% accuracy.
• Specialized Skill Pack: Provides ready-to-use scripts for ranking supernova candidates, auditing evidence, and planning nightly observation budgets.
• Academic Validation: Backed by publications in Communications Engineering and Intelligent Computing.