Your personal operating system for career transition.

Build a Type-Safe Configuration Loader for AI Services

AI services have more configuration surface than typical web backends: model names, temperature, token limits, provider API keys, retry counts, timeout values, rate limits, and feature flags. Passing these as raw environment variables or loose dicts means typos fail silently, missing required values are discovered at runtime, and the available configuration is not documented anywhere.

What to build

Implement a ServiceConfig Pydantic settings model and a load_config factory function:

ServiceConfig fields (all from environment variables with AI_ prefix):

• provider: str — required, one of "openai", "anthropic", "local"
• model: str — required, minimum length 1
• api_key: str — required for non-local providers (validate this cross-field)
• max_tokens: int — default 1024, between 1 and 32768
• temperature: float — default 0.7, between 0.0 and 2.0
• timeout_s: float — default 30.0, minimum 1.0
• max_retries: int — default 3, between 0 and 10
• max_concurrent: int — default 5, between 1 and 50
• cache_ttl_seconds: int — default 3600, minimum 0

load_config(env: dict | None = None) -> ServiceConfig — loads config from the provided dict (or os.environ if None). Raises ConfigError (a custom exception subclassing ValueError) with a descriptive message if required fields are missing or invalid.

Why this matters

Configuration bugs in AI services show up in the worst possible way: the service starts fine, then fails on the first request because the API key is wrong, the timeout is too short, or the model name has a typo. Pydantic settings validation catches all of these at startup, not at request time.

Constraints

• Use Pydantic v2 field validators, not v1 @validator.
• The api_key should be required when provider != "local" — implement this as a @model_validator.
• ConfigError should include a clear message listing what is wrong.