config.py

"""
Central configuration for the University Admissions RAG Chatbot.
All parameters are environment-variable-driven with sensible defaults.
"""

import os
from dataclasses import dataclass, field
from dotenv import load_dotenv

load_dotenv()


@dataclass
class ModelConfig:
    model_id: str = "ibm-granite/granite-4.1-3b-instruct"
    max_new_tokens: int = 512
    temperature: float = 0.3
    top_p: float = 0.9
    repetition_penalty: float = 1.15
    do_sample: bool = True
    device_map: str = "auto"
    torch_dtype: str = "auto"  # resolved at runtime


@dataclass
class EmbeddingConfig:
    model_name: str = "sentence-transformers/all-MiniLM-L6-v2"
    batch_size: int = 64
    normalize_embeddings: bool = True


@dataclass
class ChunkingConfig:
    chunk_size: int = 600
    chunk_overlap: int = 80
    separators: list = field(default_factory=lambda: ["\n\n", "\n", ". ", " ", ""])


@dataclass
class RetrievalConfig:
    top_k: int = 4
    score_threshold: float = 0.0  # cosine similarity lower-bound (FAISS L2 inverse)
    index_path: str = "faiss_index"
    index_file: str = "faiss_index/index.faiss"
    metadata_file: str = "faiss_index/metadata.pkl"


@dataclass
class AppConfig:
    title: str = "🎓 University Admissions Assistant"
    description: str = (
        "Ask me anything about admissions requirements, programs, fees, "
        "scholarships, deadlines, and more. Powered by IBM Granite 4.1 3B Instruct."
    )
    max_file_size_mb: int = 20
    allowed_extensions: tuple = (".pdf", ".docx", ".txt")
    data_dir: str = "data"
    log_dir: str = "logs"
    max_history_turns: int = 6  # pairs kept in context


@dataclass
class Config:
    model: ModelConfig = field(default_factory=ModelConfig)
    embedding: EmbeddingConfig = field(default_factory=EmbeddingConfig)
    chunking: ChunkingConfig = field(default_factory=ChunkingConfig)
    retrieval: RetrievalConfig = field(default_factory=RetrievalConfig)
    app: AppConfig = field(default_factory=AppConfig)

    @property
    def hf_token(self) -> str | None:
        return os.environ.get("HF_TOKEN")

    def validate(self) -> None:
        if not self.hf_token:
            raise EnvironmentError(
                "HF_TOKEN environment variable is not set. "
                "Export it or add it to your .env file."
            )


# Singleton
cfg = Config()