What started as a simple prompt exercise after an AWS machine learning course became SageCompass, an agentic LangGraph-based decision framework. The author, lacking prior Python experience, aimed to build a tool that could determine if a business problem warranted ML investment.

Early versions relied on prompt engineering, but quickly outgrew that approach. After hitting performance and scaling issues, the project pivoted to a multi-agent Python runtime using LangChain. Each iteration exposed brittle design choices, forcing constant restructuring.

Now in version 6, SageCompass enforces architectural contracts, integrates RAG from Drupal, and includes guardrails for deterministic behavior. The platform tackles ambiguity head-on, treating unclear input as an orchestration challenge rather than a UX failure.

Future posts will dive into repo architecture, RAG integration, and how the system handles ambiguity. The project reflects a growing trend: turning AI uncertainty into structured, testable workflows.