SynapCores AIDB - Complete Features
For Marketing and Sales Use
Last Updated: October 2025 Version: Production Ready (v2.0)
Executive Summary
SynapCores AIDB is the world's first AI-Native Database that unifies SQL, Vector Search, and Machine Learning in a single platform. Unlike traditional databases that require external tools and complex integrations, SynapCores embeds AI capabilities directly into the database engine, enabling developers to build intelligent applications with simple SQL queries.
Key Differentiators:
- ✅ SQL + Vector + ML in one database
- ✅ Native embedding generation - no external services required
- ✅ AutoML with SQL syntax - train models with a single query
- ✅ Natural language queries - ask questions in plain English
- ✅ Built in Rust - memory-safe and blazingly fast
- ✅ Multi-tenant by design - enterprise-grade isolation
1. SQL Database Features
1.1 Complete SQL Support
Standard SQL Operations:
- ✅ CREATE, DROP, ALTER tables with full DDL support
- ✅ INSERT, UPDATE, DELETE, SELECT with full DML support
- ✅ Complex JOIN operations (INNER, LEFT, RIGHT, FULL, CROSS)
- ✅ Subqueries and nested queries
- ✅ Common Table Expressions (CTEs) with WITH clause
- ✅ Recursive CTEs for hierarchical data
- ✅ UNION, INTERSECT, EXCEPT set operations
- ✅ Window functions (ROW_NUMBER, RANK, LAG, LEAD, etc.)
- ✅ GROUP BY with HAVING clauses
- ✅ ORDER BY with multiple columns and directions
Data Types:
- Standard: INTEGER, BIGINT, REAL, DOUBLE, TEXT, VARCHAR, CHAR, BOOLEAN
- Advanced: JSON, JSONB, UUID, TIMESTAMP, DATE, TIME, DECIMAL, BYTEA
- AI-Native: VECTOR(dimensions), AUDIO, VIDEO, IMAGE, PDF
Constraints & Indexing:
- ✅ PRIMARY KEY, FOREIGN KEY, UNIQUE, NOT NULL, CHECK constraints
- ✅ DEFAULT values and AUTO INCREMENT
- ✅ B-tree indexes for standard columns
- ✅ Vector indexes (HNSW) for similarity search
- ✅ Composite indexes across multiple columns
1.2 Advanced SQL Features
Stored Procedures:
- ✅ CREATE PROCEDURE with IN/OUT/INOUT parameters
- ✅ Control flow: IF/THEN/ELSE, WHILE loops, FOR loops
- ✅ Variable declarations and assignments
- ✅ Error handling with RAISE statements
- ✅ HTTP requests to external APIs
- ✅ Shell command execution (with security controls)
- ✅ Procedure overloading by parameter types
Triggers:
- ✅ BEFORE/AFTER INSERT/UPDATE/DELETE triggers
- ✅ Row-level and statement-level triggers
- ✅ Access to NEW and OLD row values
- ✅ Conditional triggers with WHEN clauses
- ✅ Trigger chaining and execution ordering
- ✅ Persistent trigger storage with multi-tenant isolation
Views:
- ✅ CREATE VIEW for virtual tables
- ✅ Updatable views
- ✅ View dependency tracking
- ✅ Persistent view definitions
Transactions:
- ✅ BEGIN, COMMIT, ROLLBACK support
- ✅ ACID compliance with MVCC
- ✅ Isolation levels: Read Committed, Repeatable Read, Serializable
- ✅ Savepoints for partial rollbacks
Window Functions (Production Ready):
- ✅ ROW_NUMBER(), RANK(), DENSE_RANK(), NTILE()
- ✅ LAG(), LEAD() for accessing adjacent rows
- ✅ FIRST_VALUE(), LAST_VALUE(), NTH_VALUE()
- ✅ Running aggregates: SUM, AVG, COUNT, MIN, MAX
- ✅ PARTITION BY and ORDER BY support
- ✅ Custom frame specifications (ROWS, RANGE)
Statistical Functions:
- ✅ STDDEV(), VARIANCE() for statistical analysis
- ✅ PERCENTILE_CONT(), PERCENTILE_DISC() for percentiles
- ✅ CORR(), COVAR_POP(), COVAR_SAMP() for correlations
- ✅ REGR_* functions for linear regression
2. Vector Database Capabilities
2.1 Vector Storage & Retrieval
Vector Data Type:
- ✅ Native VECTOR(dimensions) data type in SQL
- ✅ Support for dimensions up to 4096
- ✅ Optimized contiguous memory storage
- ✅ Automatic normalization options
Distance Metrics:
- ✅ Cosine Similarity - for text and semantic search (recommended)
- ✅ Euclidean Distance (L2) - for spatial data and image embeddings
- ✅ Dot Product / Inner Product - for recommendation systems
- ✅ Manhattan Distance (L1) - for high-dimensional sparse data
Vector Operators in SQL:
-- Cosine distance
embedding <=> query_vector
-- Euclidean distance
embedding <-> query_vector
-- Inner product
embedding <#> query_vector
2.2 Vector Indexing
HNSW Index (Approximate Nearest Neighbor):
- ✅ Graph-based index structure for fast similarity search
- ✅ Configurable parameters: M (connections), ef_construction, ef_search
- ✅ O(log n) search complexity
- ✅ 10-100x faster than brute force for large datasets
- ✅ Tunable trade-off between speed and accuracy
Flat Index (Exact Nearest Neighbor):
- ✅ Brute-force exact search
- ✅ 100% recall accuracy
- ✅ Ideal for small datasets (< 10K vectors)
Performance:
- ✅ Search 1M vectors in ~30 microseconds (HNSW)
- ✅ 128d vectors: ~0.5ms per search
- ✅ 512d vectors: ~2ms per search
- ✅ Batch operations: 10,000+ insertions/second
2.3 Embedding Generation
Built-in Models (No External API Required):
- ✅ MiniLM (384d) - Fast, general-purpose (recommended)
- ✅ BERT Base (768d) - High-quality semantic embeddings
- ✅ BERT Large (1024d) - Maximum quality for critical applications
Embedding Functions in SQL:
-- Generate embedding from text
SELECT EMBED('wireless headphones');
SELECT EMBED(product_description);
-- Specify model
SELECT EMBED(text, 'minilm');
SELECT EMBED(text, 'bert-base');
Features:
- ✅ Local inference - no external API calls
- ✅ Model caching for fast startup
- ✅ Batch processing for high throughput
- ✅ GPU acceleration support (CUDA)
- ✅ Automatic text tokenization
2.4 Semantic Search
Similarity Functions:
-- Find similar products
SELECT product_name,
COSINE_SIMILARITY(embedding, EMBED('laptop')) as score
FROM products
WHERE COSINE_SIMILARITY(embedding, EMBED('laptop')) > 0.7
ORDER BY score DESC
LIMIT 10;
Features:
- ✅ Sub-second response times
- ✅ Metadata filtering during search
- ✅ Hybrid search (vector + keyword)
- ✅ Multi-vector queries
- ✅ Threshold-based filtering
3. AI & ML Functions
3.1 Natural Language Processing
Text Analysis:
-- Sentiment analysis
SELECT SENTIMENT_ANALYSIS(review_text) FROM reviews;
-- Extract named entities
SELECT EXTRACT_ENTITIES(document_text) FROM documents;
-- Keyword extraction
SELECT EXTRACT_KEYWORDS(article_content, 5) FROM articles;
-- Text summarization
SELECT SUMMARIZE(long_text, 200) FROM articles;
Text Generation:
-- Generate text with AI
SELECT GENERATE('Write a product description for'||product_name, options)
FROM product_templates;
3.2 Multimedia AI Functions
Audio Processing:
-- Transcribe audio to text
SELECT TRANSCRIBE(audio_file) FROM recordings;
-- Get audio duration
SELECT DURATION(audio_file) FROM media;
Video Processing:
-- Extract frames from video
SELECT EXTRACT_FRAMES(video_file, 1000) FROM videos;
-- Extract audio track
SELECT EXTRACT_AUDIO(video_file) FROM videos;
-- Get video metadata
SELECT DIMENSIONS(video_file) FROM videos;
Image Processing:
-- Extract text from images (OCR)
SELECT EXTRACT_TEXT(image_file) FROM scanned_documents;
-- Get image dimensions
SELECT DIMENSIONS(image_file) FROM photos;
-- Generate image embeddings
SELECT EMBED(image_file) FROM photos;
PDF Processing:
-- Extract text from PDF
SELECT EXTRACT_TEXT(pdf_document) FROM contracts;
3.3 Vector Operations
Vector Algebra:
-- Vector arithmetic
SELECT VECTOR_ADD(vec1, vec2);
SELECT VECTOR_SUBTRACT(vec1, vec2);
SELECT VECTOR_MULTIPLY(vec, 2.5);
-- Vector analysis
SELECT VECTOR_MAGNITUDE(embedding);
SELECT VECTOR_NORMALIZE(embedding);
SELECT VECTOR_DOT(vec1, vec2);
Similarity Metrics:
SELECT COSINE_SIMILARITY(vec1, vec2);
SELECT EUCLIDEAN_DISTANCE(vec1, vec2);
SELECT INNER_PRODUCT(vec1, vec2);
4. AutoML Capabilities
4.1 SQL-Based Machine Learning
Create ML Experiments:
-- Train a classification model
CREATE EXPERIMENT churn_prediction AS
SELECT customer_id, age, tenure, monthly_charges, churned
FROM customers
WITH (
task_type='binary_classification',
target_column='churned',
max_trials=50,
algorithms=['random_forest', 'xgboost', 'neural_network']
);
Supported ML Tasks:
- ✅ Classification (binary and multi-class)
- ✅ Regression (continuous value prediction)
- ✅ Time Series Forecasting
- ✅ Clustering (unsupervised learning)
- ✅ Anomaly Detection
4.2 ML Algorithms
Available Algorithms:
- ✅ Linear Regression / Logistic Regression
- ✅ Decision Trees
- ✅ Random Forest
- ✅ Gradient Boosting
- ✅ XGBoost
- ✅ Neural Networks (multi-layer perceptron)
- ✅ K-Nearest Neighbors (KNN)
- ✅ Support Vector Machines (SVM)
- ✅ Naive Bayes
Algorithm Selection Strategies:
- ✅ All - try all available algorithms
- ✅ Fast - only fast algorithms (linear, decision trees, naive bayes, knn)
- ✅ Accurate - highly accurate algorithms (random forest, gradient boosting, xgboost, neural networks)
- ✅ Interpretable - explainable models (linear regression, logistic regression, decision trees)
- ✅ Custom List - specify exact algorithms to try
4.3 AutoML Features
Automatic Feature Engineering:
- ✅ Polynomial feature generation
- ✅ Interaction features between variables
- ✅ Automatic scaling (standard, minmax, robust)
- ✅ Missing value imputation (mean, median, mode, forward/backward fill)
- ✅ Categorical encoding (one-hot, label, target, ordinal)
Model Training:
- ✅ Automatic hyperparameter tuning
- ✅ Cross-validation (K-Fold, Stratified K-Fold, Time Series Split)
- ✅ Early stopping to prevent overfitting
- ✅ Ensemble models for better accuracy
- ✅ Configurable time budget and trial limits
Model Management:
-- View experiments
SHOW MODELS;
-- Start/stop experiments
START EXPERIMENT experiment_name;
STOP EXPERIMENT experiment_name;
-- Describe experiment results
DESCRIBE EXPERIMENT experiment_name;
-- Deploy best model
DEPLOY MODEL best_model FROM EXPERIMENT churn_prediction;
-- Make predictions
PREDICT churn_probability
USING churn_model
AS SELECT * FROM new_customers;
Performance Metrics:
- ✅ Classification: Accuracy, Precision, Recall, F1 Score, AUC
- ✅ Regression: R², MAE, MSE, RMSE, MAPE
- ✅ Clustering: Silhouette Score, Davies-Bouldin Index
- ✅ Anomaly Detection: F1 Score, Precision, Recall
5. Natural Language to SQL (NL2SQL)
5.1 Conversational Queries
Ask Questions in Plain English:
ASK "Show me all customers who made purchases last month"
ASK "What are the top 5 products by revenue?"
ASK "Find similar products to laptop computers"
Features:
- ✅ LLM-powered query understanding
- ✅ Schema-aware query generation
- ✅ Intent analysis (select, aggregate, join, time-series)
- ✅ Confidence scoring (0.0-1.0)
- ✅ Alternative query suggestions
- ✅ Query explanations in natural language
5.2 Intelligent Query Generation
Schema Context Awareness:
- ✅ Automatic table and column name inference
- ✅ Relationship detection (foreign keys, joins)
- ✅ Domain-specific terminology mapping
- ✅ Common aliases and synonyms
Query Patterns:
- ✅ Aggregation queries ("total sales by category")
- ✅ Ranking queries ("top 10 customers by spend")
- ✅ Comparison queries ("compare revenue year over year")
- ✅ Trend analysis ("sales trend over time")
- ✅ Semantic search ("find products similar to X")
Dual-Mode Processing:
- ✅ LLM-based - uses AI service for complex queries
- ✅ Pattern-based - fast rule-based generation for simple queries
- ✅ Automatic fallback if LLM unavailable