BRAID-DSPy Documentation

Welcome to the BRAID-DSPy documentation!

BRAID-DSPy is a Python library that integrates BRAID (Bounded Reasoning for Autonomous Inference and Decisions) architecture into the DSPy framework, enabling structured reasoning through Guided Reasoning Diagrams (GRD) in Mermaid format.

Quick Start

import dspy
from braid import BraidReasoning

# Configure DSPy
lm = dspy.OpenAI(model="gpt-4")
dspy.configure(lm=lm)

# Create a BRAID reasoning module
braid = BraidReasoning()

# Use it in your pipeline
result = braid(problem="Solve: If a train travels 120 km in 2 hours, what is its speed?")
print(result.answer)
print(result.grd)  # View the reasoning diagram

What is BRAID?

BRAID (Bounded Reasoning for Autonomous Inference and Decisions) is a structured reasoning framework that separates planning from execution:

Planning Phase: Generate a Guided Reasoning Diagram (GRD) in Mermaid format
Execution Phase: Execute the GRD step by step to solve the problem

This separation significantly improves reliability and reduces hallucinations compared to traditional Chain-of-Thought approaches.

Key Features

Core Capabilities

Guided Reasoning Diagrams (GRD): Generate Mermaid-format flowcharts that map solution steps
Two-Phase Reasoning: Separate planning and execution phases for better reliability
DSPy Integration: Seamlessly integrates with existing DSPy modules and optimizers
Auditable Reasoning: Visualize and debug reasoning processes through GRD diagrams
Optimization Support: BRAID-aware optimizers for improving GRD quality

BRAID Protocol Features (v0.2.0+)

Numerical Masking: Prevent answer leakage by masking computed values
Node Atomicity: Enforce ≤15 tokens per node for optimal nano-model performance
Procedural Scaffolding: Ensure GRDs describe HOW to solve, not WHAT the answer is
Stateful Execution: Dynamic GRD traversal with conditional branching
Critic Feedback Loops: Self-verification with retry mechanisms
PPD Metrics: Performance-per-Dollar analysis for cost optimization
Training Utilities: Generate synthetic data for fine-tuning Architect models

Documentation Contents

Installation

pip install braid-dspy