Skip to main content
Glama
fhir_graphrag_config.aws.yaml4.44 kB
# AWS EC2 Configuration for FHIR GraphRAG # This configuration connects to IRIS running on AWS EC2 g5.xlarge # Database Connection Settings (AWS EC2) database: iris: host: "3.84.250.46" # AWS EC2 public IP port: 1972 # IRIS SQL port namespace: "%SYS" # Use %SYS namespace (DEMO has access restrictions on AWS) username: "_SYSTEM" # Database username password: "SYS" # SuperUser password connection_timeout: 30 pool_size: 5 max_overflow: 10 # BYOT Storage Configuration (use existing vector table on AWS) # Note: Using SQLUser.ClinicalNoteVectors which already has vectorized data storage: iris: table_name: "SQLUser.ClinicalNoteVectors" column_mapping: id_column: "ID" text_column: "TextContent" metadata_columns: - "ResourceID" - "PatientID" - "DocumentType" zero_copy: true preserve_schema: true validate_table_name: false # Disable validation since this is a custom table # ✅ CloudConfiguration API vector settings (iris-vector-rag v0.5.4+) # These settings are read by SchemaManager via CloudConfiguration # Using 1024-dim for NVIDIA NIM embeddings (NV-EmbedQA-E5-v5) vector_dimension: 1024 # Vector dimensionality for NVIDIA NIM embeddings distance_metric: "COSINE" # Distance metric for similarity search index_type: "HNSW" # Vector index type # Vector Storage Configuration (AWS tables) vector_storage: table_name: "SQLUser.ClinicalNoteVectors" reference_column: "ResourceID" vector_column: "Embedding" # VECTOR(DOUBLE, 1024) model_column: "EmbeddingModel" dimension: 1024 # Legacy setting (CloudConfiguration uses storage.vector_dimension) # Knowledge Graph Storage (AWS tables) # Note: Using SQLUser schema with fully qualified names for AWS IRIS knowledge_graph: entities_table: "SQLUser.Entities" relationships_table: "SQLUser.EntityRelationships" # Embedding Configuration (NVIDIA NIM) embeddings: model: "nvidia/nv-embedqa-e5-v5" # NVIDIA NIM embedding model dimension: 1024 # Legacy setting (CloudConfiguration uses storage.vector_dimension) batch_size: 32 normalize: true device: "cuda" # Use GPU on EC2 g5.xlarge # GraphRAG Pipeline Configuration pipelines: graphrag: entity_extraction_enabled: true entity_types: - "SYMPTOM" - "CONDITION" - "MEDICATION" - "PROCEDURE" - "BODY_PART" - "TEMPORAL" relationship_types: - "TREATS" - "CAUSES" - "LOCATED_IN" - "CO_OCCURS_WITH" - "PRECEDES" min_entity_confidence: 0.7 min_relationship_confidence: 0.6 # Multi-modal search settings vector_k: 30 text_k: 30 graph_k: 10 rrf_k: 60 fusion_method: "rrf" # Performance settings batch_size: 10 parallel_extraction: true max_workers: 4 # LLM Configuration (NVIDIA NIM LLM) llm: provider: "nvidia_nim" model: "nvidia/llama-3.1-nemotron-70b-instruct" base_url: "http://3.84.250.46:8000" # NIM LLM service on AWS temperature: 0.0 max_tokens: 500 timeout: 30 fallback_to_regex: true # Logging Configuration logging: level: "INFO" format: "json" file: "logs/fhir_graphrag_aws.log" rotation: "daily" max_bytes: 10485760 backup_count: 7 # Monitoring monitoring: enabled: true metrics: - "entity_extraction_time" - "entity_extraction_count" - "relationship_extraction_count" - "query_latency" - "graph_traversal_depth" - "gpu_utilization" # AWS-specific: monitor GPU usage - "network_latency" # AWS-specific: monitor network latency performance_targets: entity_extraction_time_ms: 1000 # Faster with GPU query_latency_ms: 500 # Target sub-second queries knowledge_graph_build_time_ms: 60000 # < 1 minute for 51 documents # Feature Flags features: entity_normalization: false temporal_analysis: false entity_feedback: false query_history: true # Enable for production monitoring gpu_acceleration: true # AWS-specific: use GPU for embeddings # AWS-Specific Settings aws: region: "us-east-1" instance_id: "i-0432eba10b98c4949" instance_type: "g5.xlarge" gpu_type: "NVIDIA A10G" enable_cloudwatch: false # CloudWatch metrics integration (future)

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