We built CosmicWatch, an open-source AI system that analyzes astronomical radio-signal data and identifies unusual candidates that may deserve human investigation.
Radio telescopes generate huge amounts of data, making manual inspection difficult. CosmicWatch uses a lightweight CNN for signal detection, Isolation Forest for anomaly detection, candidate scoring for prioritization, and an open-weight LLM for explaining why a candidate was flagged.
CosmicWatch does not claim to detect aliens. Instead, it identifies unusual radio-signal patterns that researchers can investigate further, with SETI as one of its key applications.
Tech: Python, PyTorch, FastAPI, React, Tailwind, BLIMPY, Isolation Forest
GitHub:
COSMICWATCH
Open-source AI astronomical anomaly detection and candidate-prioritization system for radio SETI and Breakthrough Listen observations.
Team
Team Name: CosmicWatch Triage
Member
Contribution
Vijay (vj450-cpu)
Lead AI Engineer & Full-Stack Architect: Ingestion pipeline, dual-layer AI, candidate engine, FastAPI & React observatory dashboard
⚠️ CRITICAL SCIENTIFIC & ETHICAL MANDATECosmicWatch is NOT an "alien detector". Never claim that the system detects extraterrestrial life. The system detects unusual statistical anomalies/candidate signals and recommends candidates for human radio telescope investigation. All performance metrics originate from actual held-out empirical experiments (
$N=200$ , seed:7777). We do not invent accuracy, astronomical objects, or false extraterrestrial claims.
Scientific Taxonomy & Distinctions
To ensure scientific honesty and rigor, CosmicWatch explicitly separates:
-
Observed Astronomical Data: Authentic radio observations from public Breakthrough Listen archives (
.filand.h5files) ingested viaBLIMPY. - Synthetic Demonstration Data: Controlled artificial signal injections (narrowband, Doppler drift, intermittent carriers, bursts, and…
Built for Hacktoberfest Hack Day Coimbatore 2026.
We don't claim to have found aliens. We built a system that helps humans decide what not to ignore.
Top comments (0)