Cameron Pfiffer - Beyond JSON: Grammar-Constrained Function Calling with CFGs

Cameron Pfiffer is a Developer Relations Engineer for.txt. Cameron has been an active participant of the generative AI scene for two years, working on RAG, knowledge graph, structured generation projects. He received his PhD in financial economics from the University of Oregon, and completed a postdoctoral research position at Stanford University.

RSVP

Beyond JSON: Grammar-Constrained Function Calling with CFGs

Function calling is essential for modern AI systems, but traditional JSON approaches can be token-inefficient. This talk demonstrates how Pythonic function syntax with grammar constraints can reduce token generation by up to 2.6x while enhancing reliability.

I'll show how context-free grammars (CFGs) ensure syntactic validity and enforce business logic constraints, eliminating common errors in parameter generation. Through practical examples, we'll explore implementation strategies and compare performance between JSON and Pythonic approaches.