A comprehensive Snowflake-native application for detecting, tracking, and analyzing paranormal activity using Snowflake Cortex AI.

SnowGhost Breakers is a complete paranormal investigation system that demonstrates the power of Snowflake's data platform combined with Cortex AI capabilities. The application handles ghost sightings, evidence analysis, investigation management, and AI-powered insights.

https://www.snowghostbreakers.com/

• 📊 Comprehensive Data Model: Tables for ghosts, sightings, evidence, investigations, and sensor readings
• 🤖 Cortex AI Integration:
  • Sentiment analysis of sighting reports
  • Automated threat assessments
  • Ghost classification using LLMs
  • AI-powered image embeddings (1024-dimensional vectors)
  • Vector similarity search with custom cosine similarity
  • 💬 SnowBreakers Chat: Conversational AI interface for natural language querying
  • AI-generated investigation reports
  • Image and audio analysis with Cortex Vision AI
• 🕸️ Neo4j Graph Analytics: Network analysis of ghost relationships, community detection, centrality scoring, and predictive modeling
• 📈 Analytics & Dashboards: Pre-built semantic views and analytics
• 🎨 Streamlit Application: Interactive web interface with image upload, AI analysis, location picker, and vocabulary browser
• 📓 Jupyter Notebooks: Multimodal analytics and standard data loading pipelines
• 🔍 Cortex Analyst: Natural language queries for ghost data
• ⚡ Stored Procedures: Automated processing workflows
• 🗺️ Geospatial Analysis: Hotspot detection, mapping, and interactive sighting visualization
• 🤖 Agentic AI System: Autonomous agents for threat detection, pattern analysis, and response coordination
• 🔌 MCP Server: Model Context Protocol integration for external AI agents
• 📚 Business Vocabulary: Ghost ontology and taxonomy with searchable Streamlit interface

SnowGhostBreakers/
├── sql/
│   ├── 01_setup_database.sql          # Database and schema setup
│   ├── 02_create_tables.sql           # Core table definitions
│   ├── 03_sample_data.sql             # Sample ghost data
│   ├── 04_stored_procedures.sql       # Business logic procedures
│   ├── 05_semantic_views.sql          # Analytics views
│   ├── 06_cortex_ai_functions.sql     # AI integration examples
│   ├── 14_image_embeddings_table.sql  # **NEW: Image similarity search**
│   └── ...                            # Additional setup scripts
├── streamlit_app/
│   └── ghost_detection_app.py         # Main Streamlit application
│       └── **NEW: 🔍 Image Similarity page**
├── notebooks/
│   ├── 01_ghost_analytics.ipynb       # Analytics notebook
│   └── 02_data_loader.ipynb           # Bulk data loading
├── scripts/
│   ├── bulk_ghost_processor.py        # Bulk data processor (key pair auth)
│   └── ghost_analytics.py             # Advanced analytics (key pair auth)
├── cortex_analyst/
│   └── ghost_semantic_model.yaml      # Cortex Analyst configuration
├── setup.sql                          # Master setup script
└── README.md                          # This file

• Snowflake account with Cortex AI enabled
• Appropriate warehouse and compute resources
• Permissions to create databases, tables, and procedures

1. Clone or download this repository

2. Run the setup script in your Snowflake worksheet:

   -- Option 1: Run individual scripts in order
   !source sql/01_setup_database.sql
   !source sql/02_create_tables.sql
   !source sql/03_sample_data.sql
   !source sql/04_stored_procedures.sql
   !source sql/05_semantic_views.sql
   !source sql/06_cortex_ai_functions.sql
   
   -- Option 2: Use master setup script
   !source setup.sql

3. Verify installation:

   USE DATABASE GHOST_DETECTION;
   SHOW TABLES IN SCHEMA APP;
   SHOW VIEWS IN SCHEMA ANALYTICS;

1. Navigate to Snowflake Streamlit:

   • Go to your Snowflake account
   • Navigate to "Streamlit" in the left sidebar
   • Create a new Streamlit app
   • Copy the contents of streamlit_app/ghost_detection_app.py
   • Run the app

2. Or use Streamlit in Snowflake (SiS):

   CREATE STREAMLIT GHOST_DETECTION.APP.GHOST_DETECTION_APP
   FROM 'streamlit_app'
   MAIN_FILE = 'ghost_detection_app.py';

Master registry of all detected paranormal entities

• ghost_id, ghost_name, ghost_type, threat_level
• manifestation_frequency, confidence_score
• status tracking (Active, Dormant, Captured, Neutralized)

Individual encounter events

• Location (with geospatial coordinates)
• Environmental conditions (temperature, EMF readings)
• Witness information
• Paranormal activity level (1-10 scale)

Multimedia evidence storage

• Images, videos, audio recordings
• Metadata and file paths
• Processing status

AI-powered analysis results

• Cortex model outputs
• Confidence scores
• Entity detection
• Recommendations

Case management

• Investigation teams
• Status tracking
• Evidence collection
• Outcomes

• Sentiment Analysis: Assess the emotional tone of sighting reports
• Classification: Automatically categorize ghost types and threat levels
• Summarization: Generate concise reports from multiple sightings
• Entity Extraction: Identify key information from descriptions

• Vector Embeddings: Convert descriptions to semantic vectors
• Similarity Search: Find related sightings and patterns
• Cortex Search: Natural language search over sighting data
• 🆕 Image Embeddings: AI-powered image similarity search
  • 1024-dimensional vectors using snowflake-arctic-embed-l-v2.0-8k
  • Custom JavaScript cosine similarity function
  • Text-to-image and image-to-image search
  • Batch embedding generation
  • Real-time similarity scoring

• Report Generation: Create comprehensive investigation reports
• Threat Assessment: AI-powered risk evaluation
• Recommendations: Tactical suggestions for investigations
• Q&A: Answer questions about ghost data

-- Get AI threat assessment for a ghost
CALL GENERATE_GHOST_REPORT('GH001');

-- Classify ghost type from description
CALL CLASSIFY_GHOST_TYPE('Shadow figure moving through walls');

-- Analyze sighting with sentiment
SELECT 
    description,
    SNOWFLAKE.CORTEX.SENTIMENT(description) as fear_level
FROM GHOST_SIGHTINGS;

-- Find similar sightings
SELECT * FROM VW_SIMILAR_SIGHTINGS 
WHERE base_sighting_id = 'SIGHT001';

-- 🆕 Image similarity search
CALL FIND_SIMILAR_IMAGES('ghost orb', 10);
CALL FIND_SIMILAR_TO_IMAGE('EMB_ABC123', 5);
CALL BATCH_GENERATE_EMBEDDINGS(100);

1. VW_GHOST_ACTIVITY_SUMMARY: Comprehensive ghost activity metrics
2. VW_PARANORMAL_HOTSPOTS: Geographic concentration analysis
3. VW_INVESTIGATION_METRICS: Case performance indicators
4. VW_THREAT_MATRIX: Threat level distribution
5. VW_ACTIVITY_TIMELINE: Temporal pattern analysis
6. VW_AI_MODEL_METRICS: AI performance tracking

• Total ghost count by type and threat level
• Sighting frequency and patterns
• Geographic hotspot identification
• Investigation success rates
• AI model confidence scores
• Environmental correlations

1. 📊 Dashboard: Overview with key metrics and charts
2. 👻 Ghost Registry: Detailed ghost profiles with AI reports
3. 📍 Sightings: Location-based sighting browser
4. 🔬 Evidence Analysis: Evidence processing status and AI results
5. 📋 Investigations: Case management interface
6. 🤖 AI Insights: Interactive AI query interface
7. ➕ New Sighting: Report submission form with AI classification
8. 📈 Analytics: Advanced analytics and visualizations

• Real-time data filtering
• Geographic mapping
• AI-powered report generation
• Threat level visualizations
• Timeline analysis

The semantic model (cortex_analyst/ghost_semantic_model.yaml) enables natural language queries:

Example Questions:

• "What are the most active ghosts in the past month?"
• "Which locations have the highest paranormal activity?"
• "Show me all extreme threat level ghosts"
• "What is the average EMF reading across all sightings?"

Comprehensive analysis notebook including:

• Data overview and statistics
• Ghost type analysis
• Temporal patterns
• Geographic hotspots
• Environmental correlations
• AI model performance
• Predictive analytics

INSERT INTO GHOSTS (ghost_id, ghost_name, ghost_type, ...)
VALUES ('GH999', 'New Ghost', 'New_Type', ...);

-- Custom view example
CREATE VIEW MY_CUSTOM_ANALYSIS AS
SELECT 
    g.ghost_name,
    COUNT(s.sighting_id) as my_metric,
    SNOWFLAKE.CORTEX.COMPLETE('mistral-large2', 
        'Analyze this ghost: ' || g.description
    ) as ai_insight
FROM GHOSTS g
JOIN GHOST_SIGHTINGS s ON g.ghost_id = s.ghost_id
GROUP BY g.ghost_id, g.ghost_name, g.description;

Add new pages by modifying ghost_detection_app.py and following the existing page patterns.

• Implement row-level security for sensitive ghost data
• Use Snowflake's Dynamic Data Masking for PII (witness information)
• Configure proper role-based access control (RBAC)
• Audit trail through AUDIT_LOG table

SELECT * FROM VW_REAL_TIME_THREAT_ASSESSMENT
WHERE threat_level = 'Extreme';

SELECT 
    location_name,
    AVG(paranormal_activity_level) as avg_activity,
    COUNT(*) as sightings_last_30days
FROM GHOST_SIGHTINGS
WHERE sighting_datetime >= DATEADD(day, -30, CURRENT_TIMESTAMP())
GROUP BY location_name
HAVING COUNT(*) >= 5
ORDER BY avg_activity DESC;

SELECT 
    priority,
    COUNT(*) as case_count,
    AVG(evidence_count) as avg_evidence
FROM INVESTIGATIONS
WHERE status IN ('Open', 'In_Progress')
GROUP BY priority;

This application demonstrates:

• Snowflake table design and relationships
• Cortex AI functions (Complete, Sentiment, Embeddings)
• Stored procedure development
• Streamlit app development in Snowflake
• Semantic modeling for Cortex Analyst
• Advanced SQL analytics
• Geospatial data handling
• Time series analysis

This is a demonstration application. To extend it:

1. Fork the repository
2. Add new features or analyses
3. Test with sample data
4. Submit pull requests

Apache 2.0 License - See LICENSE file

• Inspired by the Ghostbusters franchise
• Built on Snowflake's powerful data platform
• Leverages Cortex AI capabilities
• Based on the original AIM-Ghosts project by tspannhw

Comprehensive documentation is available in the docs/ directory:

• Getting Started: Installation guides, quick starts, and tutorials
• Feature Guides: Detailed documentation for each feature
• Troubleshooting: Fix guides and error resolution
• Technical Reference: API docs, architecture diagrams

See docs/README.md for the complete documentation index.

Quick Links:

• Quick Start Guide
• Installation Guide
• Image Embeddings Fixes
• SnowBreakers Chat Guide
• Troubleshooting

Ghosts

CREATE OR REPLACE SECRET GHOST_DETECTION.APP.SLACK_WEBHOOK_SECRET
    TYPE = GENERIC_STRING
    SECRET_STRING = 'letters and numbers/it/stuff';
  
-- https://select.dev/posts/snowflake-slack-alerts


CREATE OR REPLACE NOTIFICATION INTEGRATION ghost_slack_alerts
TYPE = WEBHOOK
ENABLED = TRUE
WEBHOOK_URL='https://hooks.slack.com/services/SNOWFLAKE_WEBHOOK_SECRET'
WEBHOOK_SECRET = GHOST_DETECTION.APP.SLACK_WEBHOOK_SECRET
WEBHOOK_BODY_TEMPLATE='{"text": "SNOWFLAKE_WEBHOOK_MESSAGE"}'
WEBHOOK_HEADERS=('Content-Type'='application/json');

describe notification integration ghost_slack_alerts;

SELECT * FROM TABLE(INFORMATION_SCHEMA.NOTIFICATION_HISTORY());

CREATE OR REPLACE ALERT GHOST_TRAIL_DEMO_ALERT
WAREHOUSE = INGEST
SCHEDULE = '15 minute'
IF (EXISTS (
SELECT 1 FROM GHOST_DETECTION.APP.TRAIL_EVENTS
WHERE RECORD_TYPE = 'LOG'
AND RECORD['severity_text'] IN ('WARN','ERROR')
AND TIMESTAMP >= DATEADD(hour, -1, CURRENT_TIMESTAMP())
))
THEN CALL SYSTEM$SEND_SNOWFLAKE_NOTIFICATION(
  SNOWFLAKE.NOTIFICATION.TEXT_PLAIN(
    SNOWFLAKE.NOTIFICATION.SANITIZE_WEBHOOK_CONTENT('my message')
  ),
  SNOWFLAKE.NOTIFICATION.INTEGRATION('ghost_slack_alerts')
);



SELECT  * 
FROM GHOST_DETECTION.APP.TRAIL_EVENTS
WHERE RECORD_TYPE = 'LOG'
AND RECORD['severity_text'] IN ('WARN','ERROR')
AND TIMESTAMP >= DATEADD(hour, -1, CURRENT_TIMESTAMP());



CALL SYSTEM$SEND_SNOWFLAKE_NOTIFICATION(
  SNOWFLAKE.NOTIFICATION.TEXT_PLAIN(
    SNOWFLAKE.NOTIFICATION.SANITIZE_WEBHOOK_CONTENT('my message')),
  SNOWFLAKE.NOTIFICATION.INTEGRATION('ghost_slack_alerts')
);


SELECT
    DATE_TRUNC('day', START_TIME) AS usage_date,
    USERNAME,
    SUM(REQUEST_COUNT) AS total_messages,
    SUM(CREDITS) AS total_credits
FROM SNOWFLAKE.ACCOUNT_USAGE.CORTEX_ANALYST_USAGE_HISTORY
WHERE START_TIME >= DATEADD('day', -7, CURRENT_TIMESTAMP())
GROUP BY 1, 2
ORDER BY total_credits DESC, usage_date desc;


select *
FROM GHOST_DETECTION.APP.GHOST_SIGHTINGS;

alter table GHOST_DETECTION.APP.GHOST_SIGHTINGS
add column investigation_status varchar(25000);

{
  "image_type": "paranormal_sighting",
  "primary_subject": "ghostly_apparition",
  "description": {
    "entity": "pale_female_figure_with_long_white_hair",
    "appearance": "translucent_ethereal_being",
    "hair": "flowing_white_hair_with_motion_blur",
    "face": "pale_ghostly_complexion_looking_downward",
    "position": "emerging_from_or_through_ornate_mirror"
  },
  "environment": {
    "setting": "dark_interior_room",
    "wall_color": "dark_green",
    "focal_object": "ornate_baroque_style_mirror",
    "frame_details": "elaborate_gold_decorative_frame_with_scrollwork"
  },
  "paranormal_characteristics": {
    "transparency": "semi_translucent",
    "motion": "hair_appears_to_be_moving",
    "lighting": "supernatural_glow_around_figure",
    "classic_ghost_traits": ["white_coloring", "ethereal_quality", "mirror_manifestation"]
  },
  "atmosphere": "eerie_haunting_ominous",
  "likely_explanation": "staged_photo_or_artistic_creation_using_long_exposure_double_exposure_or_digital_manipulation"
}

• https://medium.com/@tspann/ghosts-are-unstructured-data-i-e31b34c0d9e4
• https://github.com/tspannhw/AIM-Ghosts/
• https://github.com/tspannhw/FLaNK-python-processors/tree/main
• https://github.com/tspannhw/TrafficAI

For questions or issues:

• First, check the docs/ directory for comprehensive guides
• Review Snowflake documentation for Cortex AI
• Examine the SQL comments for implementation details
• Check the sample data for usage patterns

• The application uses real paranormal investigation terminology
• EMF (Electromagnetic Field) readings are actual ghost hunting metrics
• Temperature drops are commonly associated with paranormal activity
• The 1-10 paranormal activity scale is a simplified PKE meter reading

Who you gonna call? SnowGhost Breakers! 👻🚫

Powered by Snowflake Cortex AI Tim Spann - Prompt Engineer, AI Hacker, Senior Solution Engineer, Paranormal Investigator

🎉👻✨