AI Systems & Cognitive Engines

This is my playground for experimenting with machine learning models, autonomous reasoning tools, and intelligent pipelines.

Model Playground

Astreaus-Alpha

Local Lightweight Model

Embedded Inference & Offline Reasoning

7B Parameters, 4-bit Quantized

ACTIVE

Atlas-32B

Large External LLM

Research & Hypothesis Testing

32B Parameters, FP16

CLOUD LINKED

Vision-Embedder v2

Image-to-Embedding Model

Similarity Search & Classification

ViT-L/14, 768-dim embeddings

ACTIVE

Cognitive Agent "MIND"

Dialogue & Reasoning Orchestrator

Identity-Aligned Cognition

Hybrid (Local + Cloud), Multi-model

EXPERIMENTAL

AI Experiments & Research Sandbox

Embedding Space Visualizer

Displays clusters of semantic embeddings of text or code.

Prompt-Shaping Engine

Tests and refines system prompts for optimal model performance.

Memory Routing / Context Graph

Tests hierarchical memory loading vs. linear context stitching.

Local Model Efficiency Framework

Testing quantization and CPU-only inference speeds for edge devices.

Few-Shot Prompting

Investigating the impact of n-shot examples on model accuracy and bias.

Chain-of-Thought Scaffolding

Developing structured prompts that guide the model through complex reasoning steps.

Retrieval-Augmented Generation

Integrating external knowledge bases to reduce hallucinations and improve factual accuracy.

Hallucination Detection

Building classifiers to identify and flag model-generated content that is not factually grounded.

AI Philosophy & Approach

Augmentation, Not Replacement

AI should be an augmentation of human reasoning, not a replacement of it. The goal is to build tools that amplify our own intelligence and creativity.

Models as Tools

Models are tools — intelligence emerges from the interaction with context. The true power of AI lies in the thoughtful orchestration of these tools to solve complex problems.

Optimization Beyond Accuracy

Optimization is not just about accuracy — it’s about transparency and controllability. A model that is 99% accurate but cannot explain its reasoning is a black box; a model that is 95% accurate and transparent is a partner.

The Importance of the Interface

The interface is the bridge between human and machine intelligence. A well-designed interface can make the difference between a powerful tool and an unusable one.

Interactive Model Demo

Model Graph Visualization

User InputPrompt LayerModelEmbeddingsMemoryOutput

Example Outputs

Prompt:

How many different protocols exist for IoT networking?

Model Reasoning:

The model enumerated 12 core protocols, grouping them by reliability and power constraints. It then selected the most common protocol from each category to provide a concise and representative answer.

Response:

There are numerous protocols for IoT networking, but they can be broadly categorized into several key families. The most prominent include: 1. **MQTT:** A lightweight messaging protocol ideal for low-power devices. 2. **CoAP:** A protocol designed for simple, constrained devices. 3. **HTTP/2:** A more robust protocol for devices with more processing power. 4. **LoRaWAN:** A long-range, low-power protocol for wide-area networks.

Prompt:

Generate a Python function to calculate the Fibonacci sequence.

Model Reasoning:

The model generated a recursive Python function to calculate the Fibonacci sequence. It included a base case for the first two numbers in the sequence and a recursive step for all subsequent numbers.

Response:

def fibonacci(n):
    if n <= 1:
        return n
    else:
        return fibonacci(n-1) + fibonacci(n-2)