Learning in Drublic

No BS AI development: practical Ruby tutorials that actually work.

Leadership 5 min read

AI Needs Its MVC Moment

Prompt engineering is just programming. Treat it that way and AI systems start working like real software.

V

Vicente Reig

Fractional Engineering Lead

Article 8 min read

MIPROv2 Paper: How Stanford's Prompt Optimization Works in Ruby

Deep dive into the MIPROv2 paper (arXiv:2406.11695) from Stanford. Learn how Bayesian optimization, dataset summarization, and instruction bootstrapping combine to improve LLM prompts automatically. Ruby implementation included.

V

Vicente Reig

Fractional Engineering Lead

Tutorial 4 min read

Let the Model Write Your Tools

Building a research agent with CodeAct where the LLM generates Ruby code on the fly

V

Vicente Reig

Fractional Engineering Lead

Workflow 5 min read

Evaluator Loops in Ruby: Ship Sales Pitches with Confidence

A quick walkthrough of how to turn LLM calls into composable building blocks in Ruby. With evaluator loops, you get cheap iterations, clear critiques, and real observability into each step. Great for shipping better sales pitches without guessing what the model is doing (or overspending on tokens).

V

Vicente Reig

Fractional Engineering Lead

Workflow 4 min read

Build a Workflow Router in Ruby

Route every ticket to the right Language Model, only escalate to heavy LLMs when needed, keep every hop observable, and never touch a handwritten prompt along the way.

V

Vicente Reig

Fractional Engineering Lead

Article 3 min read

TOON vs CSV for LLM Prompts: Why Nested Data Needs a New Format

CSV breaks when you need nested relationships in LLM prompts. TOON (Token-Oriented Object Notation) preserves Ruby structs, arrays, and enums while cutting token costs. See the benchmark.

V

Vicente Reig Rincon de Arellano

Fractional Engineering Lead

Article 4 min read

Cut Prompt Tokens in Half with BAML + TOON

DSPy.rb now pairs BAML schemas with Sorbet::Toon payloads. The combo keeps Enhanced Prompting simple while saving ~9000 schema tokens and ~2400 data tokens per request.

V

Vicente Reig Rincon de Arellano

Fractional Engineering Lead

Article

Rich Signatures, Lean Schemas

When signatures hit 5+ fields, JSON Schema overhead eats hundreds of tokens per call. BAML keeps them compact—no retraining needed.

V

Vicente Reig Rincon de Arellano

Fractional Engineering Lead

Article

JSON Mode Comparison: Enhanced Prompting vs Structured Outputs

Head-to-head benchmark of 8 LLM models: GPT-4, Claude 3.5, Gemini 2.0. Compare enhanced prompting vs native JSON mode for reliability, speed, and cost. Real-world performance data inside.

V

Vicente Reig

Fractional Engineering Lead

Architecture 12 min read

DSPy.rb Concurrent Architecture: Deep Dive into 3.3x Performance Gains

An architecture designed for high throughput on a fiber-based concurrent execution model, event-driven telemetry system, and production-ready architecture delivering real performance improvements with minimal overhead.

V

Vicente Reig

Fractional Engineering Lead

Production 8 min read

Observability in Action: Langfuse Tracing

See how DSPy.rb's executor-driven telemetry keeps real-time visibility without slowing down your LLM workflows

V

Vicente Reig

Fractional Engineering Lead

Features 4 min read

Google Gemini Integration: Multimodal AI with Type-Safe Ruby

DSPy.rb v0.20.0 introduces full Google Gemini support, bringing Google's state-of-the-art multimodal AI capabilities to Ruby developers with complete type safety and structured outputs.

V

Vicente Reig

Fractional Engineering Lead

Features 3 min read

Fiber-Local LM Contexts: Clean Multi-Model Management in Ruby

DSPy.rb v0.20.0 introduces DSPy.with_lm for elegant temporary language model overrides using Ruby's fiber-local storage, enabling clean concurrent patterns and better model management.

V

Vicente Reig

Fractional Engineering Lead

Features 2 min read

Run LLMs Locally with Ollama and Type-Safe Ruby

DSPy.rb now supports Ollama, bringing type-safe structured outputs to local LLM development. Learn how to build cost-effective AI applications with zero API charges during development.

V

Vicente Reig

Fractional Engineering Lead

Article

Raw Chat API for Benchmarking and Migration

Learn how to use DSPy.rb's raw_chat API for benchmarking monolithic prompts and migrating to modular implementations

D

Vicente Reig

Fractional Engineering Lead

Research 12 min read

Program of Thought: The Missing Link Between Reasoning and Code

Deep dive into Program of Thought (PoT) - a powerful approach that separates reasoning from computation. Compare it with CodeAct and ChainOfThought to find the right tool for your AI applications.

V

Vicente Reig

Fractional Engineering Lead

Tutorial 12 min read

Building Your First ReAct Agent in Ruby

Step-by-step guide to creating tool-using AI agents with DSPy.rb. Learn how to build agents that can reason about their actions and solve complex multi-step problems.

V

Vicente Reig

Fractional Engineering Lead

Design 8 min read

Building Ruby-Idiomatic AI Applications with DSPy.rb

How DSPy.rb embraces Ruby conventions to make AI development feel natural. Learn about the design decisions that make DSPy.rb uniquely Ruby.

V

Vicente Reig

Fractional Engineering Lead