Medical GraphRAG Assistant

# 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: "localhost"              # Local on EC2 (Docker container)
    port: 32782                    # IRIS SQL port (Docker mapped)
    namespace: "%SYS"              # Use %SYS namespace for reliable access 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)
knowledge_graph:
  entities_table: "RAG.Entities"
  relationships_table: "RAG.EntityRelationships"

# Multimodal Embedding Configuration (NV-CLIP)
nvclip:
  base_url: "http://13.216.2.41:8002/v1" 
  model: "nvidia/nvclip"
  batch_size: 16
  timeout: 60

# Embedding Configuration (NVIDIA NIM)
embeddings:
  model: "nvidia/nv-embedqa-e5-v5"  # NVIDIA NIM embedding model
  base_url: "http://localhost:8001/v1"
  dimension: 1024                   # Legacy setting (CloudConfiguration uses storage.vector_dimension)
  batch_size: 32
  normalize: true
  device: "cuda"                    # Use GPU on EC2 g5.xlarge

# LLM Configuration (NVIDIA NIM LLM)
llm:
  provider: "nvidia_nim"
  model: "meta/llama-3.1-8b-instruct"
  base_url: "http://localhost:8001/v1"    # NIM LLM service (local on EC2)
  temperature: 0.0
  max_tokens: 1000
  timeout: 60

# 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)