Nick

Built on Python 3.11+ because of its excellent async support, rich ecosystem, and developer experience. Considered Go for performance, but Python's ecosystem for LLM APIs is unmatched. Node.js was tempting, but Python's async/await is more mature. Perfect balance of performance and productivity.

Using Docker for deployment instead of manual setup or Kubernetes because it ensures consistency across environments and makes scaling trivial. Manual deployment was error-prone ("works on my machine"), Kubernetes was overkill for our scale. Docker Compose gave us the perfect balance. One image works everywhere.

Using Redis for caching and idempotency instead of in-memory solutions (like Python's `functools.lru_cache`) because it's persistent, scalable, and handles concurrent requests perfectly. Tried Memcached, but Redis's data structures (sets, hashes) were perfect for idempotency keys. Essential for production reliability.

Chose FastAPI over Flask/Django because of its automatic OpenAPI generation, async support, and type hints. Flask was too simple for our needs, Django too heavy. FastAPI's type hints caught bugs at development time, not production. Perfect for building production-ready APIs fast.