UnifyWeaver

Book 13: Semantic Search

Graph RAG, Embeddings, and Vector Databases

Part of the UnifyWeaver Education Series

This book covers building intelligent semantic agents using UnifyWeaver’s Python semantic runtime. You’ll learn to combine declarative logic with modern AI capabilities like vector search, graph RAG, and LLM integration.

Prerequisites

Required:

• Book 1: Foundations

Recommended:

• Book 5: Python Target - Python code generation basics

Technical:

• Python 3.8+ with pip
• Basic understanding of embeddings and vector similarity (helpful)

What You’ll Learn

• Building semantic agents with the Python runtime
• Graph-based Retrieval Augmented Generation (RAG)
• Vector search with ONNX embeddings
• Web crawling and content extraction
• LLM integration (Gemini, GPT, Claude)
• Semantic data pipelines

Chapter Overview

Chapter 1: Introduction

• Why Python for semantic tasks
• Architecture of the semantic runtime
• Your first semantic script
• Key components: PtImporter, PtCrawler, PtSearcher

Chapter 2: Graph RAG

• Graph-based knowledge retrieval
• Combining vector search with graph traversal
• Building knowledge graphs from crawled content
• RAG patterns with UnifyWeaver

Chapter 3: Semantic Data Pipeline

• End-to-end semantic pipelines
• Data ingestion and embedding
• Storage with SQLite
• Query and retrieval patterns

Chapter 4: Logic and Recursion

• Recursive semantic queries
• Transitive closure over knowledge graphs
• Combining Prolog logic with vector similarity
• Advanced query patterns

Chapter 5: Semantic Playbook

• Production deployment patterns
• Performance optimization
• Error handling and fallbacks
• Integration with other targets

The Semantic Runtime

The Python semantic runtime provides:

Component	Purpose
PtImporter	SQLite storage for objects, links, vectors
PtCrawler	XML/HTML streaming crawler with link extraction
PtSearcher	Vector search and graph traversal
LLMProvider	Wraps LLM CLIs for RAG tasks
OnnxEmbeddingProvider	Local embeddings via ONNX

Quick Start

:- module(semantic_search, [search/2]).
:- use_module(unifyweaver(targets/python_target)).

% Define a semantic search predicate
search(Query, Result) :-
    embed(Query, Vector),
    vector_search(Vector, 5, Matches),
    member(Result, Matches).

% Compile to Python
?- compile_python_semantic(search/2, [], Code),
   write_file('search.py', Code).

Installation

# Install Python dependencies
pip install numpy onnxruntime sqlite3

# Optional: Install embedding model
pip install sentence-transformers

# Optional: Install LLM CLI
pip install google-generativeai  # for Gemini

Use Cases

• Knowledge Base Search: Build searchable knowledge bases from documents
• Focused Crawling: Semantically-guided web crawling
• Document Q&A: RAG-based question answering
• Content Recommendation: Similarity-based content discovery
• Research Assistants: AI-powered research tools

What’s Next?

After completing this book, you’ll be able to:

• ✅ Build semantic search applications
• ✅ Implement Graph RAG patterns
• ✅ Create intelligent crawlers
• ✅ Integrate LLMs with declarative logic
• ✅ Deploy production semantic pipelines

License

This educational content is licensed under CC BY 4.0. Code examples are dual-licensed under MIT OR Apache-2.0.

This site is open source. Improve this page.