config.yaml

# Quickscene Production Configuration
# As specified in PRD requirements

# ASR (Automatic Speech Recognition) Configuration
asr_mode: "whisper"  # Options: "whisper", "assemblyai"
whisper_model: "tiny"  # Options: "tiny", "base", "small", "medium", "large"

# Embedding Configuration
embedding_model: "all-MiniLM-L6-v2"  # SentenceTransformer model
embedding_dimension: 384  # Expected embedding dimension

# Chunking Configuration
chunk_duration_sec: 15  # Duration of each transcript chunk in seconds
chunk_overlap_sec: 2    # Overlap between chunks to avoid cutting words

# File Paths (as per PRD structure)
video_path: "./data/videos/"
transcript_path: "./data/transcripts/"
chunks_path: "./data/chunks/"
embedding_path: "./data/embeddings/"
index_path: "./data/index/"

# FAISS Index Configuration
faiss_index_path: "./data/index/quickscene.index"
metadata_path: "./data/index/metadata.json"
index_type: "IndexFlatIP"  # Inner Product for cosine similarity

# Performance Configuration
max_workers: 4  # Number of parallel workers for processing
batch_size: 32  # Batch size for embedding generation

# Query Configuration
default_top_k: 5  # Default number of results to return
similarity_threshold: 0.3  # Minimum similarity score for results

# Logging Configuration
log_level: "INFO"  # Options: "DEBUG", "INFO", "WARNING", "ERROR"
log_file: "./logs/quickscene.log"

# Video Processing Configuration
supported_formats: [".mp4", ".avi", ".mov", ".mkv", ".webm"]
max_video_duration_hours: 5  # Maximum video duration to process

# Production Settings
enable_caching: true  # Cache embeddings and transcripts
cache_ttl_hours: 24   # Time to live for cached data
enable_monitoring: true  # Enable performance monitoring