Skip to main content
Glama
constitution.md4.86 kB
<!-- Sync Impact Report ================== Version change: (new) → 1.0.0 Added sections: - Core Principles (5 principles) - Technology Constraints - Development Workflow - Governance Modified principles: N/A (initial version) Removed sections: N/A (initial version) Templates requiring updates: N/A (initial setup) Follow-up TODOs: None --> # Medical GraphRAG Assistant Constitution ## Core Principles ### I. Authorship & Attribution All artifacts, code, documentation, and intellectual property in this project are authored and owned exclusively by **Thomas Dyar (thomas.dyar@intersystems.com)**. **ABSOLUTE PROHIBITION**: Attribution to AI coding assistants (Claude Code, Copilot, GitHub Copilot, ChatGPT, or any other AI tool) MUST NEVER appear anywhere in this project including: - README files or any documentation - Commit messages (no "Co-Authored-By: Claude", no "Generated with Claude Code") - Code comments - License files - Pull request descriptions - Any public-facing or internal materials This prohibition applies to all forms of AI attribution including emoji markers, links to AI tools, or any text suggesting AI involvement. Rationale: Thomas Dyar is the sole author and owner of all project deliverables. No AI attribution is permitted under any circumstances. ### II. MCP-First Architecture All AI assistant capabilities MUST be exposed through Model Context Protocol (MCP) tools. Direct API integrations that bypass MCP are prohibited for user-facing features. - Every search capability MUST have a corresponding MCP tool - Tool responses MUST be structured for LLM consumption - Tools MUST be independently testable via stdio transport Rationale: Ensures consistent agentic behavior and enables multiple client integrations (Claude Desktop, Streamlit, VSCode Continue). ### III. Vector Database Purity InterSystems IRIS is the single source of truth for all vector storage. External vector databases (Pinecone, Weaviate, ChromaDB, SQLite) MUST NOT be introduced. - All embeddings stored as VECTOR(DOUBLE, N) native types - VECTOR_COSINE similarity for all searches - No hybrid storage patterns mixing IRIS with other vector stores Rationale: Simplifies architecture, leverages IRIS native capabilities, reduces operational complexity. ### IV. Medical Data Integrity All medical data handling MUST preserve clinical accuracy and traceability. - FHIR resources MUST NOT be modified during ingestion - Hex-encoded clinical notes MUST be decoded faithfully - DICOM metadata MUST be preserved alongside image vectors - Entity extraction MUST use validated medical ontologies Rationale: Medical applications require auditability and data integrity for clinical decision support. ### V. Graceful Degradation System components MUST fail gracefully when dependencies are unavailable. - Missing NVIDIA NIM endpoints MUST NOT crash the application - Database connection failures MUST return informative errors - API rate limits MUST trigger backoff, not exceptions - Max iteration limits MUST prevent infinite tool loops Rationale: Production medical systems require resilience; partial functionality is preferable to complete failure. ## Technology Constraints **Required Stack:** - Database: InterSystems IRIS Community Edition with native vector support - LLM: AWS Bedrock (Claude Sonnet 4.5) - Embeddings: NVIDIA NV-CLIP (1024-dim multimodal) via NIM - Protocol: Model Context Protocol (MCP) for tool exposure - UI: Streamlit for interactive interfaces - Deployment: AWS EC2 with NVIDIA GPU (g5.xlarge minimum) **Prohibited:** - External vector databases - Non-MCP tool integrations for user features - Hardcoded credentials in source files - Synchronous blocking calls without timeouts ## Development Workflow **Code Changes:** 1. All changes MUST be tested against AWS deployment 2. Configuration MUST support both local and AWS environments via YAML 3. Scripts MUST be idempotent (safe to run multiple times) **Documentation:** 1. README MUST reflect current version and capabilities 2. PROGRESS.md tracks development history 3. Archive old implementations to archive/ directory **Testing:** 1. Integration tests MUST verify IRIS connectivity 2. MCP tools MUST have stdio transport tests 3. Embeddings MUST be validated for non-zero magnitude ## Governance This constitution supersedes all other development practices for this project. Amendments require: 1. Clear justification documented in commit message 2. Version increment following semantic versioning: - MAJOR: Principle removal or incompatible redefinition - MINOR: New principle or section added - PATCH: Clarifications and refinements 3. Update to dependent templates if principle changes affect them All code reviews MUST verify compliance with these principles. **Version**: 1.0.0 | **Ratified**: 2025-12-09 | **Last Amended**: 2025-12-09

Latest Blog Posts

MCP directory API

We provide all the information about MCP servers via our MCP API.

curl -X GET 'https://glama.ai/api/mcp/v1/servers/isc-tdyar/medical-graphrag-assistant'

If you have feedback or need assistance with the MCP directory API, please join our Discord server