← Project Archive

Knowledge systems · 2025–Present

Playbook

A multi-tenant knowledge platform that turns operational documentation into cited answers, governed agent tools, and maintainable business workflows.

Open the live product ↗

One enforcement path

I designed and developed Playbook end to end, from the initial prototype through its production architecture and shared domain core. The browser, in-product assistant, REST API, and external MCP clients all use the same backend services for tenant resolution, permissions, document lifecycle, and audit behavior—preventing each interface from becoming its own security boundary.

Grounded retrieval

  • Publishing creates a version, audit event, and tenant-scoped document chunks.
  • Hybrid search blends semantic and keyword rankings instead of betting on one retrieval method.
  • Embedding failures degrade to keyword search rather than breaking access to knowledge.
  • The assistant reads retrieved documents, returns citations, and refuses unsupported answers.

Governed agent access

  • Microsoft Entra establishes identity; Playbook membership selects the tenant and permissions.
  • MCP tools call the backend instead of reading or writing the database directly.
  • Suggestions and approvals are the default path for agent-authored changes.
  • Personal API tokens and OAuth clients inherit the same user-level controls.

Production platform

Flask API and FastMCP services run as separate Azure Container Apps, backed by tenant-partitioned Cosmos DB, private Blob Storage, and Application Insights. Repository adapters keep the domain portable across in-memory tests, local SQLite development, and Azure production, while Terraform and Azure Pipelines make the environment reproducible.

Outcome

Playbook gives teams one governed source for operational knowledge and makes that knowledge useful wherever work happens: through search, a cited chat assistant, or external AI agents. The architecture supports immediate document workflows while leaving clear seams for new tools, storage adapters, and client integrations.

Technical range

Python, Flask, FastMCP, Cosmos DB, SQLite, hybrid retrieval, embeddings, Azure OpenAI, Microsoft Entra, OAuth 2.1, Azure Container Apps, Blob Storage, Application Insights, Terraform, Docker, and Azure Pipelines.