config.py

"""
Central configuration for the DSA RAG Chatbot.
No magic numbers should be hardcoded anywhere else in the codebase — import from here.
"""

import os
from dotenv import load_dotenv

load_dotenv()

BASE_DIR = os.path.dirname(os.path.abspath(__file__))

# ---------------------------------------------------------------------------
# Flask / App
# ---------------------------------------------------------------------------
SECRET_KEY = os.getenv("SECRET_KEY", "dev-secret-key-change-me")
SESSION_COOKIE_NAME = "dsa_chatbot_session"
SESSION_LIFETIME_MINUTES = int(os.getenv("SESSION_LIFETIME_MINUTES", 60 * 24 * 7))  # 7 days

# ---------------------------------------------------------------------------
# Database (SQLite via SQLAlchemy)
# ---------------------------------------------------------------------------
DATABASE_PATH = os.path.join(BASE_DIR, "database", "app.db")
SQLALCHEMY_DATABASE_URI = f"sqlite:///{DATABASE_PATH}"
SQLALCHEMY_TRACK_MODIFICATIONS = False

# ---------------------------------------------------------------------------
# RAG / Knowledge base
# ---------------------------------------------------------------------------
KNOWLEDGE_BASE_DIR = os.path.join(BASE_DIR, "rag", "knowledge_base")
TOPIC_METADATA_PATH = os.path.join(BASE_DIR, "rag", "knowledge_base", "topic_metadata.json")
CHROMA_DB_DIR = os.path.join(BASE_DIR, "rag", "chroma_db")
CHROMA_COLLECTION_NAME = "dsa_knowledge_base"

EMBEDDING_MODEL_NAME = "all-MiniLM-L6-v2"

# Sections expected inside each knowledge-base .txt file. Used by the chunker
# to split on structure rather than arbitrary fixed-size windows.
KB_SECTION_HEADERS = [
    "DEFINITION",
    "TIME_COMPLEXITY",
    "SPACE_COMPLEXITY",
    "USE_WHEN",
    "EXAMPLE",
]

# ---------------------------------------------------------------------------
# Retrieval pipeline
# ---------------------------------------------------------------------------
INITIAL_RETRIEVAL_K = int(os.getenv("INITIAL_RETRIEVAL_K", 15))
TOP_K = int(os.getenv("TOP_K", 5))
SIMILARITY_THRESHOLD = float(os.getenv("SIMILARITY_THRESHOLD", 0.45))

# ---------------------------------------------------------------------------
# Memory (mem0)
# ---------------------------------------------------------------------------
USE_MEM0 = os.getenv("USE_MEM0", "True") == "True"
MEM0_LOCAL_STORAGE_DIR = os.path.join(BASE_DIR, "memory", "mem0_store")

# ---------------------------------------------------------------------------
# History / conversation
# ---------------------------------------------------------------------------
MAX_RECENT_HISTORY_MESSAGES = int(os.getenv("MAX_RECENT_HISTORY_MESSAGES", 6))

# ---------------------------------------------------------------------------
# LLM Provider
# ---------------------------------------------------------------------------
# "local" -> Mistral-7B-Instruct-v0.2 via transformers/bitsandbytes
# "groq"  -> future Groq-hosted provider
# "openai"-> future OpenAI provider
# LLM_PROVIDER = os.getenv("LLM_PROVIDER", "local")
LLM_PROVIDER = os.getenv("LLM_PROVIDER", "groq")

LOCAL_MODEL_NAME = os.getenv("LOCAL_MODEL_NAME", "mistralai/Mistral-7B-Instruct-v0.2")
LOCAL_MODEL_MAX_NEW_TOKENS = int(os.getenv("LOCAL_MODEL_MAX_NEW_TOKENS", 512))
LOCAL_MODEL_TEMPERATURE = float(os.getenv("LOCAL_MODEL_TEMPERATURE", 0.3))
LOCAL_MODEL_LOAD_IN_4BIT = os.getenv("LOCAL_MODEL_LOAD_IN_4BIT", "True") == "True"

GROQ_API_KEY = os.getenv("GROQ_API_KEY")
GROQ_MAX_TOKENS = int(os.getenv("GROQ_MAX_TOKENS", 1024))
GROQ_MODEL_NAME = os.getenv("GROQ_MODEL_NAME", "llama-3.3-70b-versatile")

OPENAI_API_KEY = os.getenv("OPENAI_API_KEY", "")
OPENAI_MODEL_NAME = os.getenv("OPENAI_MODEL_NAME", "")

# ---------------------------------------------------------------------------
# Logging
# ---------------------------------------------------------------------------
LOG_DIR = os.path.join(BASE_DIR, "logs")
LOG_FILE_PATH = os.path.join(LOG_DIR, "app.log")
LOG_LEVEL = os.getenv("LOG_LEVEL", "INFO")