Maxime Rivest is an applied AI engineer, open-source developer, and DSPy core contributor who builds software that makes language-model applications easier to evaluate, adapt, and use. His projects range from document-processing infrastructure to local voice transcription and an interactive handwritten diary.
With a background in scientific computing, data analytics, and the R programming ecosystem, Rivest brought a programmer’s emphasis on explicit interfaces and measurable results to AI development. Early projects included extracting tax information from farm invoices and building keyboard shortcuts that correct or rewrite text. His practical introduction to DSPy shows developers how to define a task, create reference examples, evaluate model performance, and optimize a less expensive model against a stronger baseline.
At AI Engineer World’s Fair 2026, Rivest explained how stable task definitions make AI systems more adaptable: developers can change models, prompts, reasoning strategies, or tools without rewriting the surrounding application. He organizes reliable AI programming around three complementary elements: instructions describing the goal, code enforcing nonnegotiable constraints, and examples demonstrating what successful behavior looks like.
- Evaluation-guided model optimization. His DSPy tutorials connect reference datasets, automatic optimization, and practical cost-performance tradeoffs. His system-prompt optimization work explores translation evaluations sensitive to meaning, informal register, and Quebec French vocabulary.
- Model-ready document context. His Attachments library converts files and other inputs into consistent text-and-image representations for language-model applications, extending his preference for reusable interfaces to real-world context preparation.
- AI interfaces beyond chat. Maivi provides local, real-time voice transcription with keyboard shortcuts, while Riddle turns a reMarkable Paper Pro into a diary that responds to handwritten input with animated handwriting. An automatically branching conversation project uses a language model to organize messages into a growing conversation tree.
