Implement LLMRouter class and all LLM provider classes to make tests pass:
Core implementation:
- Create ProviderResponse dataclass for provider returns (text, tokens_used)
- Update LLMRouter to unpack ProviderResponse objects
- Implement all 4 providers to return ProviderResponse:
* AnthropicProvider - calls Anthropic API with structured JSON prompts
* OpenAIProvider - calls OpenAI /v1/chat/completions endpoint
* GeminiProvider - calls Google Gemini generateContent API
* OllamaProvider - calls Ollama native or OpenAI-compatible endpoints
Router features:
- Validates provider at init time, raises UserError for unknown providers
- Reads API keys from ir.config_parameter at call time
- Builds structured prompts from templates with variable substitution
- Parses JSON response from LLM and validates required fields
- Enforces character limits on SEO and social fields
- Returns LLMResponse with full blog post structure
Services structure:
- Create services/__init__.py with exports
- Create models/__init__.py with exports
- Create tests/__init__.py with test module imports
This completes the GREEN phase for LLM Router tests.
Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
Nicholas Riegel
697b95a27b