LLM Fine-Tuning for Modern AI Teams: How One E-Commerce Unicorn Cut Inference Cost by 90%

Emmanuel Turlay, CEO/Founder of Airtrain AI, discusses how small, fine-tuned open-source LLMs can offer comparable performance to larger commercial models at a significantly lower cost. He covers the core concepts of choosing a base model, preparing a high-quality dataset, running fine-tuning jobs, and evaluating the tuned model. The talk highlights a case study where an e-commerce company reduced their inference costs by 90% by fine-tuning a smaller model for product categorization, achieving 94% accuracy compared to GPT-3.5's 47% in its untuned state.