Sókrates Wiki

Wise Delivery Offboarding Program

3 min read

Wise Delivery Offboarding Program

Summary

The Wise Delivery program is a high-priority initiative focused on the final offboarding of Wise through the delivery of two core technical systems: connectwise-etl and Eidos. Managed by the SOK team with a target completion date of March 27, the program involves transitioning these services to a production-ready state, ensuring clean exits and robust architectural handovers.

Details

The Wise Delivery program consists of two primary workstreams, both tracked via Linear and designed to be executed with the assistance of Claude Code. The program’s primary objective is to finalize the extraction and knowledge management infrastructure required for the Wise offboarding process.

ConnectWise-ETL Workstream

The connectwise-etl project is a data engineering pipeline designed to move data from the ConnectWise Manage API into a Delta Lake medallion architecture hosted on Microsoft Fabric.

  • Architecture: The pipeline utilizes Pydantic v2 for data validation, SparkDantic for Spark schema integration, and datamodel-code-generator to create self-healing schemas directly from OpenAPI specifications.
  • Key Deliverables:
    • ETLRunner Orchestration (SOK-83): Implementation of structured error collection and runner logic.
    • Gold Layer Finalization (SOK-84): Development of YAML-defined dimensions, surrogate keys, and complex joins within the medallion architecture.
    • Fabric Validation (SOK-85): End-to-end validation involving wheel builds and deployment to live Microsoft Fabric environments against the production API.
    • Documentation (SOK-86): Comprehensive README for long-term maintenance.

Eidos Workstream

Eidos is an Azure-hosted knowledge management service that serves as the intelligence layer for the offboarding. It integrates graph databases with modern AI embedding techniques.

  • Architecture: The system is built on Neo4j and utilizes Voyage AI embeddings. It supports the Model Context Protocol (MCP) and includes connectors for external data sources like GitHub, Confluence, and Jira.
  • Key Deliverables:
    • Infrastructure & Security (SOK-76, SOK-77, SOK-82): Replacing in-memory OAuth states with Redis, fixing OAuth flow safety (None-safety, DCR verification), and deploying to Azure Container Apps with CI/CD pipelines.
    • Connector Maturity (SOK-79): Completing the GitHub connector to support remote cloning, webhooks, and stale node cleanup within the knowledge graph.
    • Code Quality & Technical Debt (SOK-78, SOK-80, SOK-81): Removing legacy components (Memory Palace, DreamJobOrchestrator), resolving approximately 80 Pyright and 12 Ty type errors, and achieving a target test coverage of over 60%.

Execution Strategy

The program is structured to allow Claude Code to operate autonomously. Each task in the backlog (SOK-76 through SOK-86) was synthesized from project documentation to include specific architectural context and checkable acceptance criteria. This ensures that the AI agent can complete the migration and stabilization tasks without constant human intervention as the March 27 deadline approaches.

Graph View