Past
A quick tour of what’s new on the Fireworks platform.
When to reach for fine-tuning instead of prompting, the difference between supervised fine-tuning and other approaches, and how the managed workflow removes the usual infrastructure overhead.
Most vision models can describe and extract information from a receipt these days. Fewer, however, can reliably return the same clean JSON schema consistently across messy real-world images - line items, totals, tax, merchant name, etc. Even frontier AIs can struggle with this and if they don’t, they are quite expensive.
This session closes that gap. We’ll take an open vision model and fine-tune it to turn CORD receipt images into structured, typed output, all managed directly through Fireworks with no training infrastructure to stand up on your end. It’s a concrete stand-in for nearly any “read this document, give me structured data back” problem.
Full example and code will be available on GitHub.
Ask us anything about Fireworks, Fine-tuning, or just AI in general!
Speakers
Sinan Ozdemir
Head of AI Developer Education | Fireworks
Author of 10+ books on LLMs and agentic AI, founder of Kylie.ai (early agentic AI, YC-backed, acquired 2019), AI Lecturer/Author for Pearson + O'Reilly, and your friendly neighborhood ML engineer.