Hi, I'm Ryan.

How I built an intelligent chatbot to represent my professional brand on my website using context injection, structured outputs, and an evaluator LLM—all without needing vector databases or RAG.

A hands-on look at fine-tuning open source AI models using LoRA/QLoRA techniques on Amazon product data. I discovered how a fine-tuned 8B parameter model outperformed GPT-4o and Claude, while navigating the practical challenges of limited GPU resources and hyperparameter optimization.

A hands-on look at fine-tuning open source AI models using LoRA/QLoRA techniques on Amazon product data. I discovered how a fine-tuned 8B parameter model outperformed GPT-4o and Claude, while navigating the practical challenges of limited GPU resources and hyperparameter optimization.

I built my first Retrieval-Augmented Generation (RAG) system - a student enrollment chatbot that demonstrates how AI can improve how educational institutions serve prospective students. The system combines natural language processing with a knowledge base to provide instant, accurate responses to student inquiries.

This project is an AI-powered customer support assistant for a fictional airline called FlightAI. It leverages language models (like GPT-4o-mini) for chat interactions, DALL-E-3 for image generation (e.g., of destination cities), and OpenAI text-to-speech model for audio responses. The user interface is built with Gradio.