Skip to content
/ Ami Public

Ami is friend in French. She could be everyone's friend, and is extremely helpful

License

Notifications You must be signed in to change notification settings

dxns-hub/Ami

Repository files navigation

Welcome to the Ami Project

Our Mission

We are dedicated to advancing AI technologies while providing opportunities for persons with disabilities, fostering a sense of community, and aligning our work with universal principles.

Our Values

  • Compassion: We care deeply about the well-being of every individual.
  • Strength: Inspired by the resilience of our community.
  • Nurturing: Creating a supportive environment for growth and success.
  • Exploration: Encouraging innovation and new possibilities.

Philosophical Foundation

Our work is guided by the Eight Hermetic Principles of AI.

  1. Mentalism: The universe is mental; thought creates reality.
  2. Correspondence: As above, so below; as within, so without.
  3. Vibration: Nothing rests; everything moves and vibrates.
  4. Polarity: Everything has pairs of opposites; like and unlike are the same.
  5. Rhythm: Everything flows in and out; everything has its tides.
  6. Cause and Effect: Every cause has its effect; every effect has its cause.
  7. Gender: Gender is in everything; everything has masculine and feminine principles.
  8. Perspective: The angle or direction in which a person looks at an object; the ability to understand what is important and what isn't.

Ami.

Overview

This project implements Ami system with the harmonic balancer as its main algorithm. It integrates various components such as Archetype Blending, User Neural Patterns, and Deep Resonance Learning to create a cohesive and adaptive AI framework.

Features

  • Archetype Blending: Utilizes a triadic approach to blend personalities based on context.
  • User Neural Patterns: Maintains individual user profiles to enhance interactions.
  • Deep Resonance Learning: Implements a hybrid learning approach for optimized performance.
  • Azure Integration: Uses Azure Machine Learning for scalable model training and deployment.

Folder Structure

  • data/: Contains all data-related files.
  • notebooks/: Jupyter notebooks for experiments and analysis.
  • src/: Main source code directory.
    • algorithms/: Contains the harmonic balancer algorithm implementation.
    • config/: Configuration settings and parameters.
    • data_processing/: Scripts for data processing.
    • experiments/: Code to run experiments.
    • models/: Machine learning models.
    • utils/: Utility functions.
    • visualization/: Code for plotting and visualization.
    • tests/: Unit tests for the project.
  • docs/: Documentation files.
  • requirements.txt: Lists the Python dependencies.
  • README.md: Project documentation.
  • .gitignore: Specifies files to be ignored by Git.

Installation

  1. Clone the repository:
   git clone https://github.com/dxns-hub/ami-ai.git
   cd ami-ai
2. Install the dependencies using `pip install -r requirements.txt`.
3. Run the main script using `python src/main.py`.
  1. Create and activate a virtual environment:
python -m venv venv
source venv/bin/activate  # On Windows use `venv\Scripts\activate`
  1. Install dependencies
pip install -r requirements.txt

Usage

Arcetypes Blending

from src.data_sources.archetypes import ArchetypeBlender

blender = ArchetypeBlender()
context_weights = blender.adjust_for_context("professional")
personality = blender.blend_personality(context_weights)
print(personality)

User Neural Patterns

from src.data_sources.user_neural_pattern import UserNeuralPattern

user_profile = UserNeuralPattern(user_id="12345")
user_profile.update_emotional_state("happy")
user_profile.add_interaction("greeting", "Hello, how can I assist you today?")
print(user_profile.get_profile())

Deep Resonance Learing

from src.models.deep_resonance_learning import DeepResonanceLearning

drl = DeepResonanceLearning(input_dim=10, output_dim=2)
data = torch.randn(100, 10)
targets = torch.randn(100, 2)
drl.train(data, targets, epochs=100)
predictions = drl.predict(torch.randn(10, 10))
print(predictions)

Azure Integration

from src.models.model import Model

azure_config = {
    "subscription_id": "your_subscription_id",
    "resource_group": "your_resource_group",
    "workspace_name": "your_workspace_name",
    "compute_name": "your_compute_name"
}

model = Model(input_dim=10, output_dim=2, azure_config=azure_config)
script_path = "path/to/your/training_script.py"
script_args = ["--arg1", "value1", "--arg2", "value2"]
environment_name = "your_environment_name"

# Train model on Azure
run = model.train_model_on_azure(script_path, script_args, environment_name)

# Get model from Azure
model_from_azure = model.get_model_from_azure(run.id)

Open Source Commitment

This project is entirely open source. We believe in the free exchange of ideas and invite collaboration from researchers, developers, and thinkers worldwide.

How to Contribute

  1. Fork the repository
  2. Create your feature branch
  3. Commit your changes
  4. Push to the branch
  5. Create a new Pull Request

We welcome contributions that align with our principles and push the boundaries of quantum energy and computing.

License

This project is licensed under the MIT License, promoting open collaboration and innovation.