A housing bot should send simple FAQ questions to one foundation model and eviction-prevention counseling drafts to a stronger model based on request content. Which FM API integration approach implements that behavior?

Intelligent or content-based model routing inspects request characteristics and selects different foundation models for FAQ versus high-stakes eviction-prevention counseling drafts in a housing bot. FAQ intents invoke FM-A for speed and cost; counseling-draft intents invoke stronger FM-B for nuanced guidance. That FM API integration approach balances budget and quality instead of forcing one blunt model path for everything residents ask. Always calling both foundation models and billing the housing authority for duplicate completions fails cost control and adds latency without improving routing logic. Ignoring content type and picking a model at random for each request fails intentional design and produces unpredictable quality and spend. Routing every counseling draft to a classical linear regression endpoint instead of any foundation model fails by concept: regression does not draft counseling prose or answer FAQ language the bot requires. Exam caveat: content-based routing depends on reliable intent classification; when unsure, escalate to the stronger model or a human caseworker rather than gambling on the cheap path for eviction-prevention content. Apply stricter Guardrails and human review on counseling drafts regardless of which FM is chosen. Operational check: Implement intent or content classifiers before invoke; map FAQ to FM-A and counseling drafts to FM-B; measure cost and quality per route; sample misroutes weekly with housing staff; and ensure fallback behavior when classification confidence is low.