Great Attractor System

A research-oriented platform Great-Attractor for gravitational anomaly analysis and large-scale structure mapping of the Great Attractor (GA) region, based on embedding-based spectral similarity computation.

Overview

The Great Attractor is a gravitational anomaly located in the Laniakea Supercluster, approximately 150–250 Mpc from the Milky Way. This system provides tools for:

• Spectral similarity matching — Embedding-based vector search against a curated catalog of known celestial objects in the GA region.
• Observation data classification — Automatically classify celestial objects from survey datasets by matching spectral features against the standard catalog.
• Gravitational grade assignment — Assign gravitational influence grades to detected anomalies based on proximity and mass estimates.

Architecture

Great Attractor System
├── Data Ingestion         → Standard catalog construction from curated JSONL datasets
├── Embedding Pipeline     → High-dimensional vector encoding (X-dim) via embed
├── Vector Search Engine   → Milvus Lite (inner product similarity) for nearest-neighbor retrieval
├── Web Interface          → FastAPI + vanilla JS single-page application
└── Result Export          → Classification results exported as Excel/CSV

Quick Start

Start Web Service

python main.py

Specify Port

python main.py --port 8080

HTTPS Mode

python main.py --ssl-keyfile key.pem --ssl-certfile cert.pem --port 8666 --host 0.0.0.0

Dependencies

• Python >= 3.10
• FastAPI + Uvicorn (web framework)
• Milvus Lite (local vector database)
• aiohttp (async HTTP client)
• openpyxl / pandas (data I/O)
• tqdm (progress tracking)

Install dependencies:

uv sync

Project Structure

├── main.py                    # FastAPI application entry point
├── app/
│   ├── core/
│   │   ├── config.py          # Configuration (embedding endpoint, DB path, etc.)
│   │   ├── vectoring.py       # Milvus vector client & embedding retrieval
│   │   └── utils.py           # Async concurrent utilities
│   └── processors/
│       ├── standard_data_processor.py   # Build standard catalog from JSONL
│       ├── user_data_processor.py       # Classify observation data via vector search
│       └── excel_processor.py           # Excel ↔ CSV conversion
├── scripts/                   # CLI utilities for pipeline operations
├── data/
│   ├── standards/             # Standard catalog files (JSONL)
│   ├── raw/                   # Uploaded observation datasets
│   └── processed/             # Classification results
├── static/
│   └── index.html             # Web frontend
└── logs/                      # Application logs

Data Pipeline

1. Catalog Construction — Parse the standard celestial object catalog (standard.jsonl), encode each entry's spectral description into a 4096-dimensional vector, and store in a Milvus Lite collection.
2. Observation Classification — For each observation record, encode the spectral annotation, search the vector catalog for the nearest match, and append classification + grade + matched object metadata.
3. Result Export — Output classified datasets as downloadable Excel files.

Configuration

Key settings are defined in app/core/config.py:

Setting	Description
Embedding API	embed endpoint for spectral vector encoding
Milvus DB Path	Local Milvus Lite database file
Collection Name	standard_telecom (historical name, stores the celestial catalog)
Embedding Dimension	4096

License

Internal research use only.