Ai

This project builds an AI system based on Nasdanika capabilities, focusing on operating on resource collections (interconnected models). It describes model elements and their relationships from multiple angles through the 'narrator' processor, and uses embeddings and vector storage to implement semantic search and RAG (Retrieval - Augmented Generation). It also supports the chat completion functions of OpenAI and Ollama.

Knowledge management and memory Artificial intelligence chatbots #AI Model #Semantic Search #Vector Storage #Intelligent Chat .Java

rating : 2 points

downloads : 10

update time : 2025-04-29

What is the Nasdanika AI Model Interpreter?

This is an AI-based system that can automatically analyze the elements and their relationships in a data model and generate easy-to-understand explanations in natural language. The system can handle various types of model data, such as family relationships and organizational structures.

How to use the Nasdanika AI Model Interpreter?

Users only need to provide the data model. The system will automatically analyze the relationships between model elements, generate natural language descriptions, and build a semantic index for subsequent queries.

Applicable Scenarios

Suitable for scenarios that require explaining complex data relationships, such as family genealogy analysis, organizational structure description, and knowledge graph display.

Main Features

Intelligent Relationship ExplanationAutomatically identify and explain various relationships between model elements, such as parent - child and sibling relationships.

Multi - angle DescriptionGenerate multiple ways of explaining the same relationship from different angles.

Semantic SearchImplement semantic search function based on a vector database, considering semantic and graphical distances.

Multi - model SupportSupport multiple AI models such as OpenAI and Ollama.

Advantages and Limitations

Advantages

Automatically generate easy - to - understand natural language explanations

Support multi - angle descriptions of complex relationships

Intelligent search combining semantic and graphical distances

Support multiple AI model backends

Limitations

Large - scale vector databases require more storage space

The initial index construction time is long

Some technical knowledge is required for initial setup

How to Use

Prepare the Data Model

Convert your data model into a format supported by the system

Generate Explanation Text

Run the interpreter to generate natural language descriptions of model elements

Build the Vector Database

Create a semantic index for the generated explanation text

Query and Interact

Query model information through the command line or Web interface

Usage Examples

Family Relationship ExplanationThe system can automatically explain the relationships between family members

Attribute ExplanationThe system can explain the attribute definitions of model elements

Relationship QueryQuery all instances of a specific relationship

Frequently Asked Questions

What data model formats does the system support?

How long does it take to build the vector database?

How much storage space does the system require?

How to choose the AI model backend?

Related Resources

Family Relationship Example

An example showing how the system handles family relationships

Vector Database Library

The implementation of the vector database used by the system

Pre - trained Word Vectors

Download of the pre - trained word vectors used by the system

🚀 AI Operations Based on Nasdanika

This project focuses on AI-related operations, specifically those built on Nasdanika's capabilities. It abstracts low-level implementation details of AI components through models and resource sets.

🚀 Quick Start

This project is centered around AI operations, especially those leveraging Nasdanika's features. It operates on resource sets, which are collections of interconnected models. Models and resource sets abstract the low - level implementation details of AI components.

✨ Features

Narrator Processor: The "Narrator" processor can describe model elements and their relationships in various ways. For example, in the Sample Family, a model with derived relationships and ability-based reasoning can explain the following:
- Paul is Leah's father.
- Leah is Paul's daughter.
- Leah is Paul's child.
- Elias is Leah's brother.
- Elias is Leah's older brother.
- ...
- Additionally, through the EObject -> EClass relationship, it can explain that Leah is a woman and then provide the definition of a woman in the meta - model. This text can be added to Leah's description. The generated text can then be chunked, embedded, and added to the vector store for semantic search and RAG (Retrieval - Augmented Generation), considering both semantic and graph distances.

🔧 Technical Details

Key Components

Embeddings: From OpenAI and Ollams.
Vector Store: hnswlib. Observations from a local computer:
- Loading approximately 200K vectors from the pre - downloaded dl.fbaipublicfiles.com/fasttext/vectors-crawl/cc.en.300.vec.gz file takes about 10 minutes.
- The index file is approximately 2.5 Gb.
- Loading the index from the file takes about 90 seconds.
Chat Completion
CLI (Command - Line Interface):
- Vector store mix - in functionality.
- Generate embeddings for search documents in .json format.
- Create a new vector store from search documents in .json format, with or without embeddings. It can be a brand - new creation or an update to the existing content.
- Semantic search HTTP server routes.
- Chat Vuejs component - Refer to here.
- Chat with the model on a static website while providing semantic search commands, OpenAI, and Ollama. Use Docker on a GCP virtual machine.
- Have a conversation with the model.