rudrani-rane/ATIS

🛰️ ATIS — Automated Threat Intelligence System

Planetary Defense Mission Control · Near-Earth Object Monitoring · v1.0

ATIS is a research-grade planetary defense intelligence platform that ingests real NASA JPL asteroid data, runs a Graph Neural Network to learn orbital threat relationships, and surfaces results through an interactive multi-page mission-control dashboard.

Demo

Features

AI Threat Engine

• —Graph Attention Network (GAT) trained on 12,054 Near-Earth Objects
• —GNN + Gradient Boosting hybrid classifier — 5-fold cross-validated accuracy: 99.7%, Recall: 99.1%, F1: 99.2%, ROC-AUC: 0.9999
• —Orbital-only input (neo/pha columns stripped to prevent label leakage)
• —17 orbital features → 32-dimensional latent embedding space
• —Probabilistic output heads: mean (µ) + uncertainty (σ) per asteroid
• —Hybrid threat score combining GNN intelligence with physical orbital parameters (MOID, H, eccentricity, inclination)
• —PHA probability per asteroid from dedicated classifier head trained with weighted BCE loss

Live Data Integration

• —Real-time asteroid ingestion from NASA JPL Small Body Database (SBDB) API
• —Background updater polling for new close-approach events
• —WebSocket broadcast of live threat score changes to all connected clients

Dashboard Pages (Version 1.0)

Alert System

• —Auto-generates alerts from watchlist when no history is available
• —Classifies alerts as critical, high, warning, or info based on GNN threat score
• —Live update events injected every 25 seconds from top-threat asteroids

Tech Stack

Getting Started

Prerequisites

• —Python 3.10+
• —pip

Installation

git clone <repo-url>
cd "Automated Intelligence Threat System"
python -m venv venv
venv\Scripts\activate        # Windows
# source venv/bin/activate   # Linux/macOS
pip install -r requirements.txt

Run the Server

uvicorn src.web.main:app --port 8000

Open http://localhost:8000 in your browser.

Note: On first launch ATIS runs GNN inference, fits the PHA probability classifier, and computes 5-fold cross-validation scores. This takes ~60–90 seconds before the server accepts requests.

Retrain the GNN (optional)

python -m src.models.train

GNN Model Details

Scientific Formulas & Algorithms

Keplerian Orbital Mechanics

Kepler's Equation (Eccentric Anomaly):

E - e·sin(E) = M

Solved using Newton-Raphson iteration.

Position Calculation:

x = a·(cos(E) - e)
y = a·√(1-e²)·sin(E)

Velocity (Vis-Viva Equation):

v = √[μ·(2/r - 1/a)]

Impact Energy

Kinetic Energy:

E = ½·m·v²
E_MT = E / (4.184 × 10¹⁵) joules

Crater Diameter (Scaling Law):

D_crater = 1.8 · ρ_a^0.11 · ρ_t^(-1/3) · L^0.13 · v^0.44 · g^(-0.22) · sin(θ)^(1/3)

N-Body Simulation

Gravitational Acceleration:

a_i = Σ [G·m_j·(r_j - r_i) / |r_j - r_i|³]

Verlet Integration:

v(t+Δt/2) = v(t) + a(t)·Δt/2
r(t+Δt) = r(t) + v(t+Δt/2)·Δt
a(t+Δt) = F(r(t+Δt))/m
v(t+Δt) = v(t+Δt/2) + a(t+Δt)·Δt/2

Anomaly Detection

Z-Score:

z = (x - μ) / σ

Isolation Forest Score:

s(x, n) = 2^(-E(h(x))/c(n))

VERSION 2 - LATER

The following features are planned for the next major release and are not yet available in v1.0:

Pages Reserved for v2

Planned Improvements

• —Proper location-based galaxy view — real Keplerian orbital positions at J2000 epoch instead of z-score approximations
• —Automated nightly sync with full JPL SBDB catalog
• —MPC (Minor Planet Center) observation data for trajectory refinement
• —Temporal GNN — model orbital evolution over time
• —SHAP-based per-asteroid local explainability
• —Monte Carlo impact probability distributions
• —Deflection mission delta-V planner
• —User accounts with personal watchlists and email alerts
• —Docker/container packaging for one-command deployment
• —Mobile-responsive layout pass
• —AR/VR mode for the 3D galaxy view (WebXR)

Data sourced from NASA/JPL — see JPL SBDB for data terms.