Reference Architectures
Instead of designing from scratch, start from a proven architecture and adapt it. These reference patterns cover the most common AI system types — each one shows the component stack, the data flow, the key design decisions, and the failure modes to guard against before you deploy.
In This Section
RAG Chatbot Architecture
The canonical retrieval-augmented generation chatbot — ingestion pipeline, vector store, query processing, reranking, generation, and citation rendering.
Voice Pipeline Architecture
STT → VAD → LLM → TTS pipeline design — latency budgets at each stage, interruption handling, streaming output, and on-device vs cloud trade-offs.
Coding Agent Architecture
Autonomous coding agent design — codebase context, tool loop (read/write/run/test), sandboxed execution, and the human-in-the-loop approval layer.
Hybrid Router Pattern
Route queries across multiple models by task type, cost tier, and data sensitivity — classifier-based routing, fallback chains, and cost-quality trade-off measurement.