AI-powered CRM recommendation engine combining KPI metrics, competitor intelligence, and contextual data through a four-layer processing pipeline before sending structured prompts to Mistral AI for narrative generation. The architecture routes user inputs from a browser frontend through Node.js APIs, validates and calculates deltas across email marketing KPIs, injects context into French-language prompts, and normalizes AI responses into unified assisted or fallback modes. This pattern demonstrates graceful degradation—when AI services timeout or fail, the system automatically returns KPI-only analytics without breaking the user experience. Fork this diagram to customize the validation rules, add additional AI providers, or adapt the prompt engineering layer for your domain-specific metrics. The modular design separates concerns across validation, calculation, prompt construction, and normalization, making it ideal for teams scaling recommendation engines with reliability requirements.