config/epic2_graph_calibrated.yaml

# Epic 2 Calibrated Configuration - Graph Retrieval Optimized
# Graph parameters optimized through calibration system (Score: 0.8000)
# Optimal parameters: graph_weight=0.1, similarity_threshold=0.5

# Document processor for handling input files
document_processor:
  type: "hybrid_pdf"
  config:
    chunk_size: 1024
    chunk_overlap: 128

# Embedding generator for converting text to vectors  
embedder:
  type: "modular"
  config:
    model:
      type: "sentence_transformer"
      config:
        model_name: "sentence-transformers/multi-qa-MiniLM-L6-cos-v1"
        device: "mps"
        normalize_embeddings: true
    batch_processor:
      type: "dynamic"
      config:
        initial_batch_size: 64
        max_batch_size: 256
        optimize_for_memory: false
    cache:
      type: "memory"
      config:
        max_entries: 100000
        max_memory_mb: 1024

# Epic 2 ModularUnifiedRetriever with Graph Enhancement Enabled
retriever:
  type: "modular_unified"
  config:
    vector_index:
      type: "faiss"
      config:
        index_type: "IndexFlatIP"
        normalize_embeddings: true
        metric: "cosine"
    
    sparse:
      type: "bm25"
      config:
        k1: 1.2
        b: 0.75
        lowercase: true
        preserve_technical_terms: true
    
    # Epic 2 Feature: Graph Enhanced Fusion - ENABLED
    fusion:
      type: "graph_enhanced_rrf"
      config:
        k: 60
        weights:
          dense: 0.4    # Reduced to accommodate graph
          sparse: 0.3   # Maintained
          graph: 0.1    # CALIBRATED: Optimal graph weight
        graph_enabled: true
        similarity_threshold: 0.5  # CALIBRATED: Optimal similarity threshold
        max_connections_per_document: 15
        use_pagerank: true
        pagerank_damping: 0.85
    
    # Identity reranker (neural disabled for graph-only testing)
    reranker:
      type: "identity"
      config:
        enabled: false

# Answer generation strategy
answer_generator:
  type: "adaptive_modular"
  config:
    prompt_builder:
      type: "simple"
      config: {}
    llm_client:
      type: "ollama"
      config:
        model_name: "llama3.2:3b"
        base_url: "http://localhost:11434"
        timeout: 30
    response_parser:
      type: "markdown"
      config: {}
    confidence_scorer:
      type: "semantic"
      config: {}

# Global settings
global_settings:
  environment: "testing"
  log_level: "info"