Example Code Notebooks - WithPi SDK

Check out the Pi Playgrounds you want to demo some of these techniques in a UI.

Category & Description	Example Code Notebooks
Scoring System & Calibration Design evaluative frameworks that quantify and guide AI optimization toward your specific definition of “good” performance.	Using user preference logs to calibrate a scoring system
Prompt Optimization Systematically refine prompts using evaluation metrics to improve model outputs for specific tasks.	Try DSPy to optimize your prompt to perform well for your scoring system Try Pi’s own optimizer to automatically write you a prompt that works better for your scoring system
Inference Optimization Fine-tune multi-model configurations to balance quality, latency, and cost-efficiency.	Dynamically choose between a large and small model to optimize the tradeoffs between quality vs latency and cost
Model Fine Tuning Customize model weights using targeted training data to enhance performance on domain-specific tasks.	Use your scoring system to steer the GRPo Reinforcement Learning algorithm popularized by DeepSeek Use a set of example data to fine tune your own model
Data Acquisition & Processing Source, filter, and prepare high-quality datasets tailored to effectively train and evaluate models.	Import a HuggingFace dataset and filter it down to just the examples that work great according to your scoring system Synthesize a set of user prompts so you can evaluate models’ performance across diverse scenarios Synthesize a set of input/output pairs that you can use to tune a model
Retrieval Augmented Generation (RAG) Teach your model to better use data from an external database	Teach your model to break down complex queries into simple, database friendly queries

Introduction to the Pi SDK Get an API key