← BACK TO PROJECTS
The Cosmic Detective: Multi-Agent AI for Sky Exploration
#Multi-Agent AI
#LLM Orchestration
#Neo4j Knowledge Graph
#Astrophysics Data
#Swift FITS Pipeline
#Gemini API
Developed a collaborative multi-agent artificial intelligence framework within a 7-member research team for autonomous sky survey data exploration, transient detection, and astrophysical hypothesis generation.
System Architecture & Pipeline Highlights
- Multi-Agent Cognitive Architecture:
- Implemented specialized autonomous agents assigned distinct observational roles (Survey Ingestion Agent, Anomaly Verification Agent, Literature Cross-Referencing Agent, Hypothesis Synthesis Agent).
- Orchestrated fault-tolerant LLM tool-use routines utilizing Google Gemini coupled with live academic literature search and citation cross-verification.
- Semantic Knowledge Graph (Neo4j):
- Constructed a graph database in Neo4j linking observed sky coordinates, spectral profiles, and historical astronomical catalogs.
- Deployed transformer-based embedding models with a 95% semantic similarity threshold for automatic claim and observation de-duplication across hundreds of published papers.
- High-Throughput Swift FITS Pipeline:
- Engineered an optimized pipeline in Swift to stream and process heavy Flexible Image Transport System (FITS) observation files from space and ground-based telescopes.
- Extracted temporal, spectral, and photometric flux features at scale to feed downstream anomaly detection models.