Machine learning and deep learning neural networks were revolutionized by the Transformer architecture for AI models, and Large Language Models (LLMs) soon emerged, with ChatGPT exploding into the consumer marketplace. Increasingly, Generative AI features are essential to any modern product, even though the entire tech stack is still in its early days. In any emerging tech, simple, clean interfaces are essential, and a grounding in sound software engineering helps smooth out the rough patches. The OgbujiPT toolkit arose out of frustration with over-complicated and questionably engineered libraries for working with LLMs. It allows you to get started using language and multi-modal (language+visual) AI models.

OgbujiPT supports self-hosted, local models, even such lightweight/mobile friendly models as Phi-2, as well as full-service offerings such as OpenAI's GPT models. It includes support for sophisticated AI/LLM techniques such as Vector Databases, Retrieval Augmented Generation (RAG, often used for "chat your documents" tools) and having LLMs integrate with live actions and real-time function-calling. You can use it with back ends such as llama.cpp (custom HTTP API), llama-cpp-python (OpenAI HTTP API), actual ChatGPT, text-generation-webui (AKA Oobabooga or Ooba), as well as in-memory hosted Llama-class, etc. models via Python libraries such as ctransformers.

Basic outline:

• Brief overview of GenAI and LLMs
• AI language generation in five lines of Python
• Flexible usage modes for GenAI interfaces
• Vector databases and retrieval-augmented generation
• Agent capabilities in GenAI and function calling