Marie-Chatfield-Rivas-AI-Conf-Learn-Less-Parse-More_-Using-Grammars-To-Post-Process-Generated-Text-1 Learn Less, Parse More_ Using Grammars To Post-Process Generated Text”]
LLMs are trained with expansive corpora, but very little of that text includes your domain, your business rules, and your syntax. Standard output parsers can handle simple lists or JSONābut what if your requirements are more complex? Take advantage of cutting-edge models with less time spent fine-tuning, training, and prompt engineering by creating your parsers to post-process generated text.
Design your custom grammar and transform LLM responses in real-time: fix hallucinations, enforce complex validation, and translate user intent into your domain-specific language. Learn the basics of combining LLMs and parsers with three real-world examples: generating SQL queries, custom search syntax, and RDF models with a strict ontology. You will generate novel responses with probability, then deliver them with consistency and certainty.
Marie Chatfield Rivas (they/them) is a product-focused staff software engineer and experienced conference speaker who specializes in technical deep dives that are accessible for beginners and illuminating for experts. They currently build LLM-backed features for data.world, focusing on using natural language to interact with knowledge graphs beyond chat boxes.