RAG / Agent / Kafka Oral Drill

A verbal fundamentals screen covering RAG, LangGraph / LangChain agent design, tool use, MCP, Kafka consumer groups, Redis / MySQL persistence, concurrency, and long-context memory handling. Requirements 60-minute network / agent-flavored technical screen. The first half is oral fundamentals; the second half can include linked-list coding. Concrete oral topics: RAG architecture and how the candidate used it in a project. LangGraph / LangChain usage and how the agent framework is designed. Agent tool use, MCP, and how tools are exposed to the agent. Agent memory: whether the system stores memory, how memory is retrieved, and what happens when conversation history exceeds a very large context window. Kafka rebalance behavior. Kafka topics, partitions, and consumer groups; whether different consumer groups receive the same messages. Network layers. Concurrency vs. parallelism. Python GIL. Whether MySQL and Redis can substitute for each other, whether Redis can persist data permanently, and how persistence is configured. Notes The round mixes project-specific agent questions with backend fundamentals. Do not answer RAG only at buzzword level; tie retrieval, memory, and tool execution back to your actual system design. For long-context handling, prepare a concrete policy: summarize old turns, retrieve memory snippets, evict low-value context, and store durable memory outside the prompt. For Kafka, be precise: partitions are consumed by members within a consumer group; separate consumer groups each receive their own copy of the topic stream. Preparation Draw your RAG / agent architecture from memory: ingestion, embedding, retrieval, reranking, prompt assembly, tool invocation, memory write-back. Rehearse Kafka consumer group and rebalance explanations with a small partition/member example. Prepare concise answers for Redis persistence modes, MySQL-vs-Redis trade-offs, Python GIL, and concurrency vs parallelism.