galsl/semantic-search repository overview

Local File System Code Search using Hybrid Approach

A local implementation of semantic code search that operates entirely on the local filesystem. Based on code_search_s3vectors.py functionality but adapted for local storage.

Features:

1. C++ code parsing using libclang
2. Creates code chunks with metadata
3. Supports dual embedding providers:
   • AWS Bedrock (amazon.titan-embed-text-v2:0)
   • Local sentence-transformers (all-MiniLM-L6-v2 or other models)
4. Local file system storage (mirrors source tree structure)
5. Three search modes:
   • Semantic only (vector similarity)
   • Metadata only (filter-based)
   • Hybrid (filter + semantic search)
6. Two search methods with timing comparison:
   • Brute-force (linear scan)
   • FAISS index (IVF/HNSW for large datasets)
7. LLM-based query parsing (natural language → metadata filters)

Usage:

# Initialize index
python3 local_FS_search_using_hybrid_approach.py init \\
    --index ./my_code_index \\
    --embedding-provider aws  # or 'local'

# Index C++ files
python3 local_FS_search_using_hybrid_approach.py index \\
    --index ./my_code_index \\
    src/*.cpp

# Search (semantic)
python3 local_FS_search_using_hybrid_approach.py search \\
    --index ./my_code_index \\
    "authentication logic"

# Search (hybrid)
python3 local_FS_search_using_hybrid_approach.py search \\
    --index ./my_code_index \\
    --mode hybrid \\
    "bool methods in User class for validation"

# Compare brute-force vs FAISS performance
python3 local_FS_search_using_hybrid_approach.py search \\
    --index ./my_code_index \\
    --compare-methods \\
    "error handling"